Vermont has become the 23rd U.S. state to enact a comprehensive consumer privacy law after Governor Scott signed S.71, the Vermont Data Privacy and Online Surveillance Act (VDPOSA), on June 16. This new law is amongst the broadest in the country, closely resembling the 2025 version of the Connecticut Data Privacy Act (CTDPA). For example, the VDPOSA includes low applicability thresholds, a broad definition of sensitive data, heightened protections for consumer health data, consumer rights to know third parties to whom your personal data is sold and to contest certain profiling decisions, and impact assessments for certain uses of profiling. The law will take effect on January 1, 2028 and be enforced exclusively by the attorney general.
In addition to enacting the VDPOSA, Vermont also passed bills updating the state’s data broker registry (H.211), establishing a direct-to-consumer genetic testing law (H.639), and recognizing a right to neural privacy (H.814). This blog post provides background on Vermont’s privacy legislative efforts in recent years, then covers the law’s scope and key definitions, consumer rights, business obligations, and enforcement provisions. The blog post concludes with a brief overview of other privacy legislation enacted in Vermont this year.
Background
Privacy has been a long time coming in the Green Mountain State. Two years ago, Governor Scott became the first governor to veto a comprehensive consumer privacy bill. That bill, H.121, was an omnibus privacy bill with comprehensive protections and an age-appropriate design code. Had that bill been enacted, the comprehensive privacy provisions would have been amongst some of the broadest and most stringent in the country. In particular, the bill included Maryland-style substantive data minimization requirements, a ban on selling sensitive data, and a limited private right of action (PRA). The legislature tried, but failed, to overturn the veto.
The legislature continued working on privacy issues in the intervening years. Last year, they enacted the Vermont Age-Appropriate Design Code Act. This year, they finally reached consensus on a comprehensive consumer privacy law as well as an update to the state’s data broker registry, regulation of direct-to-consumer genetic testing companies, and a “right” to “mental and neural data privacy.” Although the law enacted this year diverges from the 2024 effort in notable ways, this law nevertheless incorporates many elements from the broadest and most privacy protective iterations of the Washington Privacy Act framework in the country.
Scope and Key Definitions
Covered Entities: The law applies to persons who conduct business in Vermont or produce a product or service targeted to Vermont residents and, excluding payment transaction data, annually either (1) control or process the personal data of at least 35,000 consumers, (2) control or process the sensitive data of at least 3,000 consumers, or (3) offer for sale the personal data of at least 3,000 consumers. These thresholds are low compared to those in other states, and it is uncommon to include a threshold tied to processing sensitive data. Like Connecticut’s and Maryland’s laws, the VDPOSA has requirements for consumer health data and consumer health data controllers that are not subject to the same applicability thresholds, instead applying broadly to “a person that conducts business in [Vermont] or a person that produces products or services that are targeted to residents of [Vermont.” This law also addresses any potential conflicts with the Vermont Age-Appropriate Design Code Act (AADCA), providing that the most protective law should control in any situation where that law conflicts with the requirements of this law. (Section 1, § 2415b.)
Definitions: The law’s definitions are generally consistent with the Connecticut model, including aspects of Connecticut’s 2023, 2025, and 2026 amendments. Two definitions worth noting:
“Publicly available information” is defined narrowly, excluding information collated and combined to create a consumer profile that is made available via a publicly available website and inferences generated from such profiles. For the consumer rights to access, correction, and portability (see below), publicly available information does not include “information made available for sale,” meaning such data is subject to those rights. This narrow definition of publicly available information—which conversely expands the scope of personal data—is similar to Connecticut’s updated definition from this year’s CTDPA amendments.
“Sensitive data” is defined broadly, including less common categories such as data revealing status as nonbinary or transgender, consumer health data, neural data, information derived from genetic or biometric data, certain financial account numbers or login information, or government-issued identification numbers. (Section 1, § 2415a.)
Entity and Data-Level Exemptions: The law includes many of the common entity-level exemptions, including for: certain government entities acting “in the ordinary course of its operation”; a covered entity or business associate under HIPAA (although a “hybrid entity” is not fully subject to the exemption); state or federally chartered banks or credit unions or affiliates or subsidiaries principally engaged in financial activities; certain health care providers and health care facilities under Vermont law; nonprofits established to detect and prevent insurance fraud; and more. Continuing a trend in recent years, the VDPOSA opts for more targeted entity-level exemptions for specific types of financial entities and nonprofits rather than broader exemptions for all GLBA-regulated entities and all nonprofits.
The law also includes many of the common data-level exemptions, including for: certain health records, patient identifying information, and research data; activities using information for the purpose of evaluating creditworthiness, credit standing, credit capacity, character, general reputation, personal characteristics, or mode of living if “done strictly in accordance with” the FCRA by a consumer reporting agency, furnisher, or person using a consumer report; information collected, processed, or disclosed in accordance with the DPPA or FERPA; data subject to GLBA; protected health information under HIPAA; personal data of a victim or witness of certain crimes (e.g., child abuse, human trafficking) maintained by a victim services organization; and more. (Section 1, § 2415c.)
Exceptions for Common Business Activities: The law includes many exceptions which are consistent with existing state comprehensive privacy laws, including: compliance with federal, state, or municipal laws or regulations; compliance with investigations, subpoenas, or summons; compliance with law enforcement agencies; preventing or detecting security incidents, fraud, or illegal activity; engaging in public or peer-reviewed scientific or statistical research in the public interest that meets required safeguards; internal use of data for product improvement or for internal operations reasonably aligned with the expectations of the consumer; and more. (Section 1, § 2415i.)
Consumer Rights
Consumers have the standard rights to confirm whether a controller is processing their personal data and access that data, correct inaccuracies in their personal data, delete their personal data, obtain a copy of their personal data in a portable format (if technically feasible), and to opt-out of the processing of their personal data for targeted advertising, the sale of personal data, or profiling in furtherance of a decision that produces a legal or similarly significant effect concerning the consumer. These rights contain a few unique or uncommon provisions:
The right toaccess extends to inferences about the consumer derived from personal data and the right to know whether a controller or processor is processing the consumer’s personal data for profiling to make a decision that produces legal or similarly significant effects;
Consistent with states like Minnesota and Montana, a controller is prohibited from disclosing certain categories of personal data in response to a consumer accessrequest. These include social security numbers, government-issued ID numbers, financial account numbers, health or medical identification numbers, account passwords, security questions or answers thereto, or biometric data. If a controller has collected any of these categories of personal data, it must inform the consumer “with sufficient particularity” that it has done so.
If a consumer’s personal data was processed for profiling in furtherance of an automated decision that produced any legal or similarly significant effect, the consumer has the right to, “if feasible,” question the profiling result, be informed of the reason the profiling resulted in the decision, and review the personal data processed for the profiling. If the decision concerned housing, the consumer can also correct any inaccurate personal data that was processed and have the decision reevaluated based on the corrected data. This right to contest adverse profiling decisions matches that in Connecticut’s law, which was added last year and modeled on Minnesota’s broader right.
Consumers have an additional right to obtain a list of the third parties to whom their personal data was sold, or, if the controller does not maintain such a consumer-specific list, then a list of all third parties to whom personal data is sold.
This law allows consumers to designate an authorized agent to opt out of processing on the consumer’s behalf (including for profiling) and to use an opt-out preference signal to opt out of the sale of personal data or targeted advertising. (Section 1, § 2415d.)
Business Obligations
Controllers and processors have enumerated responsibilities under the law, including transparency, data minimization, data security, oversight of processors, antidiscrimination, heightened protections for minors, and conducting both data protection assessments and impact assessments.
Transparency: Controllers must provide consumers with a “reasonably accessible, clear, and meaningful” privacy notice that includes required information under the law, such as categories of data processed, processing purposes, how to exercise rights and appeal decisions, the categories of personal data sold to third parties, and the categories of third parties to whom personal data is sold. (Section 1, § 2415e(c).)
limit the collection of personal data to what is “reasonably necessary and proportionate” in relation to the purposes for which the data are processed, as disclosed to the consumer;
obtain consent to process personal data for any “material new” purpose that is “neither reasonably necessary to, nor compatible with,” the disclosed purposes for which the personal data is processed; and
not process a consumer’s sensitive data without the consumer’s consent and “unless the processing is reasonably necessary in relation to the purposes for which the sensitive data are collected.”
Although these provisions are not “substantive data minimization” requirements in the same way that Maryland’s or California’s are, they are slightly unusual. In particular, the “necessary and proportionate” language is a departure from the usual “adequate, relevant, and reasonably necessary” language used in most state laws based on the WPA framework. Only Connecticut uses this same language, and those requirements were added in last year’s CTDPA amendments. Nevertheless, this is still a procedural requirement that ties data collection to the purposes disclosed to the consumer. Also similar to Connecticut, this law explicitly states that a controller cannot sell a consumer’s sensitive data without consent. (Section 1, § 2415e(a).)
Data Security: Controllers are required to “establish, implement, and maintain reasonable administrative, technical, and physical data security practices to protect the confidentiality, integrity, and accessibility of personal data.” (Section 1, § 2415e(a)(2).)
Processors: Controllers must engage in oversight of processors by entering into a contract that meets statutory criteria (e.g., providing instructions for processing data, describing the nature and purpose of the processing, imposing confidentiality). (Section 1, § 2415f.)
Antidiscrimination: Controllers are prohibited from processing personal data in violation of a federal or state law that prohibits unlawful discrimination against consumers. Similar to Connecticut’s 2025 amendment, this law further provides that, for state laws only, any evidence (or lack thereof) of proactive anti-bias testing or similar efforts to avoid processing data in violation of any anti-discrimination law will be relevant to any claim for a violation of such a state law. The law also includes a narrow exception for internal data use in profiling to correct bias. (Section 1, § 2415e(a)(5).)
Consumer Health Protections: Similar to Connecticut’s and Maryland’s law, this law includes heightened protections for consumer health data, such as confidentiality requirements for employee access to consumer health data, a prohibition on geofencing health care facilities for certain purposes (within 1,850 feet), and a prohibition on selling consumer health data without the consumer’s consent. Consumer health data is defined broadly as “any personal data that a controller uses to identify a consumer’s physical or mental health condition, diagnosis, or status,” and it includes reproductive or sexual health data and gender-affirming health data. These protections apply more broadly than the rest of the law to “persons,” notwithstanding the law’s other applicability thresholds. (Section 1, § 2415k.)
Assessments: Like most comprehensive privacy laws, this law requires controllers to conduct and document a data protection assessment for certain processing activities that present a heightened risk of harm to consumers, including: processing personal data for targeted advertising; selling personal data; processing personal data for profiling that presents a reasonably foreseeable risk of substantial injury to consumers, or processing sensitive data. Once again taking inspiration from Connecticut’s 2025 amendment, this law will additionally require a controller to conduct an impact assessment for any profiling conducted for making a decision that produces legal or similarly significant effect. These impact assessments must include information such as: the purpose, intended use, and deployment context of the profiling; analysis on whether the profiling presents a reasonably foreseeable risk of harm; descriptions of inputs and outputs; post-deployment monitoring and user safeguards; and more. The Vermont Attorney General (AG) may request a completed data protection or impact assessment as part of an investigation. (Section 1, § 2415g.)
Minor-Specific Provisions Address Potential Conflicts with the Vermont AADCA The VDPOSA prohibits a controller from processing personal data for targeted advertising or selling the consumer’s personal data if the controller “has actual knowledge, and willfully disregards,” that a consumer is at least 13 years of age but younger than 18 years of age. Maryland’s law includes a similar prohibition, albeit with a different knowledge standard. The VDPOSA clarifies that a controller who is also a covered business under the Vermont AADCA must comply with the requirements in that law. Where the two laws conflict, the most protective law will control. (Section 1, §§ 2415b & 2415e(a)(7), (9).)
Enforcement
The VDPOSA will be enforced exclusively by the attorney general. The law includes a permissive cure period of 60 days, allowing the attorney general to issue a cure notice to an alleged violator if the attorney general “determines that a cure is possible.” This cure period will expire on June 30, 2029. Although this law does not include a private right of action (PRA), the legislature added a statement of intent declaring that the attorney general will bear the burden of enforcing the law and, if sufficient appropriations and resources are not provided, the legislature will consider adding a PRA. (Section 1, § 2415j; Section 2; Section 3.)
Updates to Data Broker Registry Headline Other Privacy Efforts
The VDPOSA may be the most notable privacy bill enacted in Vermont this year, but it is not the only one. Vermont also updated the state’s data broker registry (H.211), enacted a direct-to-consumer genetic testing law (H.639), and established a right to neural privacy (H.814).
Data Brokers: Vermont is one of several states to create a data broker registry, alongside California, Connecticut (enacted this year), Oregon, and Texas. Effective January 1, 2027, H.211 significantly amends Vermont’s law. Key changes include—
Re-defining “brokered personal information” (“BPI”) as “any information, including derived data and unique identifiers, that is linked or reasonably linkable, alone or in combination with other information, to an identified or identifiable individual or to a device that identifies, is linked to, or is reasonably linkable to one or more identified or identifiable individuals in a household.” The prior definition was narrower and limited to an enumerated list of identifiers, plus a catch-all. This definition, in contrast, is modeled on the broader definitions of “personal data” under many state comprehensive privacy laws. Additionally, the “publicly available information” exception to the definition of BPI has been broadened. The previous definition only excluded publicly available information “to the extent that it is related to a consumer’s business or profession.” That qualifier has been removed.
Re-defining “data broker” by changing the “direct relationship” requirement. The core definition remains unchanged: a data broker is “a business . . . that knowingly collects and sells or licenses to third parties the [BPI] of a consumer with whom the business does not have a direct relationship.” Previously, the law provided examples of direct relationships, such as a “past or present . . . customer, client, subscriber, user, or registered user of the business’s goods or services.” The newly revised definition is instead modeled on California’s Delete Act regulations: a consumer must have “intentionally interacted with a business for the purpose of accessing, purchasing, using, requesting, or obtaining information about the business’s products or services,” and “[a] business is still a data broker and does not have a direct relationship with a consumer as to the [BPI] the business sells about the consumer that it collected outside of a first-party interaction with the consumer.” The revised definition also removed some of the law’s exceptions to data brokerage, such as “developing or maintaining third-party e-commerce or application platforms.”
Adding a definition of “sale” that is consistent with the common definition under state comprehensive privacy laws: the exchange of BPI for monetary or other valuable consideration, subject to exceptions.
Imposing due diligence or ‘know your customers’–style requirements prior to disclosing BPI, such as: requiring “prospective users” of BPI to identify themselves, state their purposes for seeking the information, and certifying that the information will not be used for other purposes; and, prior to disclosing such information, verifying the prospective user’s identity, reviewing the stated purposes for using the information, and not disclosing the information where there are “reasonable grounds for believing that the information will be used to violate State or federal law or will not be used for the purposes stated by the user.”
Adding new security breach notice requirements specific to data brokers.
Accelerating when a data broker must register, closing the timeframe to 30 days after a data broker first meets the relevant definition and then annually by July 1 of each year.
Increasing the registry fee and imposing a $20,000 bond.
Expanding the mandatory disclosures during registration. New information that a data broker must disclose includes whether the data broker: collects certain categories of personal information (e.g., precise geolocation data, immigration status); has sold or shared the consumers’ data, in the past year, with certain entities (e.g., foreign actors, state or federal government, “developer of a GenAI system or model”); “the URL of a page on the data broker’s website” that includes relevant information on consumer rights; and more. Notably, while this disclosure requirement references consumer rights to deletion, opting-out of the collection of BPI, and authorized agents’ ability to exercise an opt-out on the consumer’s behalf, the revised data broker law itself does not require that data brokers offer these rights. The newly enacted VDPOSA, however, will require data brokers to offer consumer rights if they are within the scope of that law.
Improving consumer access to the data broker registry in the form of a downloadable spreadsheet of registered data brokers.
Although earlier versions of H.211 would have added a California Delete Act–style accessible deletion mechanism, the final bill merely directs the Vermont Secretary of State to study the feasibility of establishing an accessible deletion mechanism.
Genetic Testing: Vermont has become the third state this year—after South Dakota and Connecticut—to enact a law regulating direct-to-consumer genetic testing. The Vermont Genetic Information Privacy Act will go into effect on July 1, 2026. This law imposes notice and consent requirements for the collection and use of biological samples and genetic data, gives consumers rights of deletion and access, and prohibits certain disclosures or uses of genetic data. Violations of the law will constitute unfair and deceptive acts in commerce under 9 V.S.A. § 2453. This includes a private right of action, although consumers will have to provide written notice of an alleged violation to a direct-to-consumer genetic testing company or service provider prior to initiating a civil action and allow 30 days to cure the notice. The cure requirement will expire on June 30, 2028. The Attorney General has enforcement and rulemaking authority.
Mental and Neural Privacy: Although the primary focus of H.814 is extending the duration and scope of the state’s Artificial Intelligence Advisory Council, this law also formally recognizes an individual right to “mental and neural data privacy.” This includes rights to “change an individual’s decision regarding neurotechnology,” to “be afforded protection from unauthorized neurotechnological alterations in mental functions critical to personality,” and to “be afforded protection from unauthorized neurotechnological alterations in mental functions critical to personality.” The law does not define key terms such as “neurotechnology,” nor are there specific mechanisms or business obligations attached to these new rights. The newly enacted VDPOSA includes neural data as a category of sensitive data, however, providing Vermont residents with actionable protections like opt-in consent requirements and mandatory data protection assessments.
Pictured: Vermont receiving its red star on the FPF “Privacy Patchwork” quilt.
Future of Privacy Forum Announces 2026 Career Achievement Award Recipients
WASHINGTON, D.C. — The Future of Privacy Forum, a global non-profit focused on data protection, AI, and emerging technologies, announced new recipients of its Career Achievement Award, recognizing exceptional leaders whose work has advanced privacy, responsible data governance, and AI leadership worldwide.
The 2026 recipients of the FPF Career Achievement Award are:
Anita L. Allen, one of the world’s foremost scholars of the law and ethics of data and technology, and Professor of Law at the University of Pennsylvania Penn Carey Law School; Henry R. Silverman Professor of Law and Professor of Philosophy, Emeritus
Nuala O’Connor, a leading voice in the responsible and ethical use of technology and data, a board member and senior advisor to companies and nonprofits, and the first Chief Privacy Officer of the Department of U.S. Homeland Security
Harriet Pearson, an esteemed technology governance and policy advisor and Managing Principal at Axia Advisory, former Cybersecurity Division Head, New York Department of Financial Services, former Partner, Hogan Lovells, first Chief Privacy Officer and Security Counsel, IBM Corporation.
Alan Raul, FPF’s Board Chair, and Jules Polonetsky, FPF CEO, were honored to introduce the awardees and presenters at the organization’s Advisory Board Annual Meeting in Washington, D.C.
“At a moment when conversations around privacy, AI, and digital governance are evolving faster than ever, these honorees represent the very best of principled leadership,” said Polonetsky. “Anita Allen, Nuala O’Connor, and Harriet Pearson have each made extraordinary contributions to advancing thoughtful, practical, and responsible approaches to data protection and emerging technologies. Their work has shaped the field in profound and lasting ways.”
“Each of this year’s honorees has helped define the modern privacy field through decades of leadership, scholarship, and public service,” added Raul. “Their work has influenced legal frameworks, corporate governance practices, academic thought, and public policy conversations around the world. FPF is proud to recognize their leadership.”
Anita L. Allen
Anita Allen is an internationally recognized scholar, philosopher, and legal theorist whose work has helped shape foundational conversations around privacy rights, ethics, civil liberties, and data governance. She is the Henry R. Silverman Professor of Law and Professor of Philosophy, Emeritus at the University of Pennsylvania Carey Law School and served as Vice Provost for Faculty. A prolific author, educator, and thought leader, Allen has examined issues of consent, accountability, equality, and ethical responsibility across both public and private life. Allen is also the author of more than 130 articles and chapters, as well as several books, including Unpopular Privacy: What Must We Hide. She is an elected member of the National Academy of Medicine, the American Law Institute, the American Philosophical Society and a fellow of the American Academy of Arts and Sciences. Throughout her distinguished career, she has advised policymakers, data protection authorities, and institutions on questions at the intersection of privacy, technology, and democratic values, helping establish frameworks that continue to influence privacy law and ethics globally.
Nuala O’Connor
Nuala O’Connor is a globally recognized leader in the responsible use of data and technology in our daily lives. She currently serves as senior advisor on technology, privacy, and AI to multinational companies Maersk and Kekst CNC and is a board member at several nonprofit organizations including EqualAI, the Society for the Rule of Law, and the Future of Privacy Forum. Nuala has held technology and legal leadership roles at General Electric, Amazon, DoubleClick, and Walmart, and served as President and CEO of the Center for Democracy & Technology (CDT). She has held senior appointed positions in the U.S. Federal government, including serving as the country’s first statutory chief privacy officer, when she was appointed as CPO of the U.S. Department of Homeland Security. Throughout her career, O’Connor has been a leading voice advancing practical, human-centered approaches to emerging technology policy, responsible innovation, and digital governance.
Harriet Pearson
Harriet Pearson is one of the pioneers of modern corporate privacy leadership and cybersecurity governance. She is the founder of Axia Advisory LLC, where she consults on program design and governance, coaches senior leaders, and supports strategic communications and external engagement. Pearson served for more than a decade as IBM’s first Chief Privacy Officer, helping establish one of the world’s most influential corporate privacy programs during a transformative period for global data governance and technology regulation. Until June 2024, Pearson was Executive Deputy Superintendent and head of the Cybersecurity Division at the New York Department of Financial Services where she led a comprehensive update of DFS’s cybersecurity regulation. Previously, Pearson was a Partner at Hogan Lovells where she founded and led the global cybersecurity practice for a decade starting in 2012. In addition to her corporate leadership, she has served in advisory and board roles across the technology, policy, and nonprofit sectors, helping organizations navigate complex questions around accountability, ethics, and responsible innovation. Pearson’s work has been at the leading edge of technology innovation and her work has led the implementation of privacy leadership as a core business and governance function at companies around the world.
FPF’s Annual Awards recognize individuals whose work has had a lasting impact on privacy protections, responsible innovation, and the advancement of ethical technology governance worldwide.
To learn more about the Future of Privacy Forum, visit fpf.org.
##
Future of Privacy Forum Releases Comprehensive Report On Algorithmic Personalization in Youth Online Experiences
As policymakers continue to debate youth online safety regulations, a new FPF report assesses the role of data-driven personalization and its implications for emerging policy and product design
WASHINGTON, D.C. — (June 10, 2026) — The Future of Privacy Forum — a global non-profit focused on data protection, AI, and emerging technologies —today released Personalization and Youth Online: Assessing Benefits, Risks, and Safeguards, a comprehensive report examining the role of algorithmic personalization in young people’s digital lives and its implications for emerging policy and product design.
Personalization—the use of personal data to tailor content and services to individual users—sits at the center of today’s most contested debates about youth and digital technology. It appears in nearly every digital service young people use—influencing the content they see, how information is ranked, which recommendations are surfaced, and what prompts or ads are presented. While it offers a host of functional and protective benefits, the practice has also become a recurring regulatory focus of efforts to protect minors from online harms, including by targeting the specific product features and design elements through which personalization operates.
“Personalization is an incredibly nuanced topic; the same mechanisms that may be associated with risks may also enable benefits such as adaptive learning, community discovery, and age-appropriate protections that make the online experience safer for young people,” said Daniel Hales, Policy Counsel for the Future of Privacy Forum and co-author of the report. “Understanding the tradeoffs that come with any risk mitigation strategy is critical, so you don’t inadvertently limit the benefits of personalization along with the harms.”
In an effort to help policymakers and companies effectively evaluate these tradeoffs, the report examines the intersection of personalization and youth online experiences in five parts: key definitions and common use cases; key benefits of personalization in youth online experiences; common risks of harm; an assessment of more than a dozen emerging mitigation proposals and the associated tradeoffs; and an analysis of the emerging policy and regulatory landscape.
Examples of the mitigation strategies examined in the report include:
Alternative approaches to curating content, such as feeds that display content chronologically, by popularity, or by categories and preferences a user has explicitly selected. While these approaches may reduce data-intensive curation, the report notes that they may have unintended effects, including the potential to inadvertently amplify spam and misinformation, and limit platforms’ ability to suppress age-inappropriate content.
Data minimization rules applied without reference to the purpose of how the data is being used may curtail beneficial personalization alongside higher-risk applications. Purpose-based limitations, which restrict data use according to processing context rather than data category alone, may better preserve the protective uses of personalization while constraining its riskier applications.
Risk mitigations related to conversational AI chatbots, including those designed to facilitate transparency, consent, responsible data practices, and safety-by-design. Each of these comes with its own tradeoffs; for example, on-device processing of sensitive data protects privacy, but can reduce model quality, which may ultimately push users toward less protective alternatives. Effective safeguards for AI personalization will require careful attention to which constraints reduce overall risk and which simply shift it.
Safety and well-being measures, including parental control and oversight, increasing digital literacy, measures to improve digital well-being, and control over personal data use. Many of these tactics will be most effective when paired with other measures; for example, increased digital literacy and making default wellbeing features like time-use limits and sleep and nighttime protections do not address the underlying design features that generate risk.
“Effective regulation must account for some precise questions: how personalization is implemented, what data it uses, and what purpose it serves,” said Holly Hawkins, Director for Youth Policy for the Future of Privacy Forum and the other co-author of the report. “While no single mitigation strategy is going to effectively address the full range of risks, we know that the most effective approaches share a common characteristic: they take a risk-proportionate approach to limiting higher-impact potential harms while preserving the functional and protective benefits of personalization for young people. We hope that this report can serve as a valuable resource to policymakers and companies who are attempting to find this delicate balance.”
The report’s release follows a growing interest from policymakers at both the federal and state levels in regulating personalization practices as part of efforts to strengthen protections for youth online. Both New York and California have passed laws prohibiting online services from providing algorithmically curated feeds to minors without parental consent; South Carolina and Nebraska’s age-appropriate design code laws require services to offer an opt-out of personalized recommender systems.
The full report, including an appendix tracking a sample of enacted and proposed youth online safety laws worldwide that address personalization, is available here.
To learn more about the Future of Privacy Forum, visit fpf.org.
###
Frontier AI Goes Federal: How the Great American AI Act Compares to State Laws
Introduction
It has been an unusually active few weeks for AI safety policy. Following a new frontier model safety bill passed in Illinois, and a White House executive order on AI security, Rep. Jay Obernolte (R-CA) and Rep. Lori Trahan (D-MA) released a bipartisan discussion draft for the Great American AI Act of 2026, adding another major federal proposal to the rapidly developing frontier AI landscape.
The draft is broad, covering issues ranging from workforce development and AI literacy to cybersecurity and international standards. But for many AI developers and deployers, the most important provisions are those focused on frontier model regulation. The draft would create requirements related to frontier AI transparency, critical safety incident reporting, employee whistleblower protections, and independent verification organizations. It would also include a three-year preemption clause restricting state laws that specifically regulate AI model development.
This blog highlights four key takeaways of the discussion draft:
The draft is one of the first bipartisan attempts in Congress to address both frontier model safety and preemption of AI. These aspects make it a notable legislative effort, even if its prospects are uncertain.
The draft incorporates many of the frontier model safety provisions in existing state laws but also has key distinctions. Compared to recent state frontier AI laws in California, New York, and Illinois (pending signature), the bill makes some important adjustments, like adding a revenue threshold for “frontier developers,” modifying the definition of “critical safety incident,” and utilizing a different penalty structure.
The draft brings the preemption debate back into the federal AI policy conversation.It includes a three-year preemption clause focused on state laws that specifically regulate AI model development.
The draft also reaches beyond frontier model safety. Other notable provisions include a study content moderation, a federal voluntary model testing program, and disclosure requirements for AI-related mass layoffs.
The Act Enters the AI Safety and Preemption Debate
Amidst a crowded but unsettled federal AI policy landscape, the Great American AI Act is notable for its regulatory focus, bipartisan backing, and what it may signal for federal AI governance ahead. Though Congress has introduced no shortage of AI bills, there has been limited movement toward enacted legislation. Congress has considered sector-specific bills on chatbots1, regulatory sandboxes, defense, elections, and financial scams, as well as broader proposals aimed at establishing a national AI framework like Sen. Blackburn’s (R-TN) TRUMP AMERICA AI Act.
Amid this activity, the Great American AI Act steps into two of the most active and contested AI policy debates: frontier model safety and federal preemption. On AI safety, the draft follows a period of growing attention to frontier model oversight at both the state and federal levels, where policymakers are trying to balance concerns about catastrophic risks and national security with concerns that overly burdensome requirements could slow AI innovation or weaken U.S. competitiveness.
On preemption, the draft arrives less than a year after Congress rejected a much broader effort to pause state AI regulation. In July 2025, the Senate voted 99-1 to remove a proposed 10-year moratorium on state AI laws from the budget reconciliation package. By contrast, the Great American AI Act includes a narrower three-year preemption provision focused on state laws that specifically regulate AI model development.
The draft is also significant because of both who introduced it and how they introduced it. Other federal AI bills have addressed AI safety or included preemption language, but this proposal comes from bipartisan sponsors who have been closely involved in federal AI policy2. In introducing the draft, the sponsors emphasized that “policy for a technology this transformative can only be built to last if it’s written by both parties.” Just as importantly, they have framed the draft as a starting point rather than a final product, describing it as “the start of a serious national conversation.” Whether the bill advances in its current form remains uncertain, especially given the sensitivity of federal preemption and the range of issues addressed. Even if the draft changes, it may help shape the terms of future federal debates over AI safety. At a minimum, it is another sign that Congress is increasingly focused on how to govern AI systems and how federal rules should interact with the fast-moving state AI landscape.
Frontier AI Requirements: Where the Draft Aligns with State Laws
The draft’s frontier model provisions are not starting from scratch. They closely track the framework emerging from California’s SB 53, New York’s RAISE Act, and Illinois’ SB 315 (awaiting signature), including requirements for frontier AI frameworks, transparency reports, safety incident reporting, enforcement, and whistleblower protections. But the federal draft makes some important adjustments, including a revenue threshold for “frontier developers,” a definition of “critical safety incident” that is broader in some respects and narrower in others, and a federal enforcement structure with penalties of up to $1 million per day.
Key Definitions
The Draft largely aligns with recent state frontier AI laws’ definitions of “large frontier developer,” “frontier model” (encompassing foundation models trained using more than 10^26 computational operations), and “catastrophic risk” (covering certain risks of death, serious injury, or major property damage arising from frontier models). For more background on SB 53 and the RAISE Act, see FPF’s prior analysis. There are, however, a few notable differences.
First, unlike the state laws, the Draft requires “frontier developers” to have a gross revenue exceeding $50 million. The state laws only include a gross revenue qualifier in the definition of “large frontier developer.”
Second, the draft’s definition of “catastrophic risk” does not specify that the death, serious injury, or property damage must arise from a “single incident,” as in the state laws.
Finally, the draft’s definition of “critical safety incident” differs from the state laws by not mandating actual harm occur (such as death, bodily injury, or property loss), nor does it include scenarios where the model uses deceptive techniques to evade developer controls or monitoring.
Core Frontier Model Obligations
The draft would impose several obligations on frontier developers and large frontier developers that also largely mirror the recent state frontier AI laws. The main requirements include:
Frontier AI Framework: Large frontier developers would need to write and publicly post a frontier AI framework addressing catastrophic-risk thresholds and mitigations, model weight cybersecurity, internal governance, and decisions about internal use and deployment. Developers would need to review the framework at least annually and publish any material modification within 30 days. These requirements closely track the state laws.
Transparency Reports: Before, or concurrently with, deploying a new frontier model or a model with a substantial modification, frontier developers would need to publish a transparency report covering information such as release date, supported languages, output modalities, intended uses, restrictions, risk assessments, third-party involvement, and mitigation steps. These requirements largely align with recent state frontier AI laws.
Reporting Mechanisms: The draft would require confidential reporting mechanisms for critical safety incidents and for catastrophic risks arising from a large frontier developer’s own use of its frontier models, including internal use and internally deployed models. This is similar to the state laws in allowing both frontier developers and members of the public to report critical safety incidents, but the draft adds a separate mechanism for reporting catastrophic risks from developers’ own model use.
Critical Safety Incident Disclosures: Frontier developers would need to report critical safety incidents to the Center for AI Standards and Innovation (CASI) within 15 days of discovery, or within 24 hours if the incident poses an imminent risk of death or serious injury. This mirrors California’s SB 53, while New York and Illinois require a shorter 72-hour reporting window.
Enforcement: Violations can result in fines of up to $1 million per violation, with each day treated as a separate violation. Federal and state attorneys general may also seek injunctions. This penalty structure may be less stringent than the state laws: SB 53 allows penalties of up to $1 million per violation, while the RAISE Act and SB 315 allow penalties of up to $3 million for subsequent violations. However, unlike the draft, those laws do not clearly cap penalties for a continuing violation at $1 million per day, meaning multiple violations could potentially be enforced in a single day.
Definitional Updates and Rulemaking: Beginning by January 1, 2028, the CAISI Director would need to issue annual recommendations on whether key definitions should be updated. The Secretary of Commerce would also have authority to adopt rules implementing the provisions, including criteria for determining when model modifications are “substantial” or “material.” California’s SB 53 includes a similar definitional update requirement, while New York’s RAISE Act similarly provides rulemaking authority.
Employee Whistleblower Protections: In a separate section, the draft would prohibit AI companies from retaliating against employees for lawfully reporting violations of federal AI laws and would provide remedies for employees who experience retaliation. This provision is similar to whistleblower protections in California’s SB 53 and Illinois’ SB 315, though those laws go further by requiring frontier developers to provide covered employees with clear notice of their rights and access to a reasonable internal process for anonymously disclosing information.
Taken together, the draft would bring much of the emerging state frontier AI framework into federal law. The broad architecture is familiar, but the federal draft contains key distinctions. It also does not carry over every state-law mechanism, most notably the frontier developer disclosure programs in the RAISE Act and SB 315, which require large frontier developers to maintain current filings with state agencies on ownership and business information.
Independent Verification Organizations and Audits
The draft would also put independent auditors at the center of its frontier AI framework. Within one year of enactment, and every six months thereafter, large frontier developers would need to retain a licensed independent verification organization (IVO) to verify compliance with the draft and assess whether the developer’s risk mitigation efforts are adequate to address catastrophic risks.
The provision reflects a broader trend toward independent evaluation of frontier models. Illinois SB 315 would also require large frontier developers to undergo independent third-party audits, but it does not create a licensing and oversight system for IVOs like the one proposed in the draft. However, other states, like Virginia, have begun exploring this model. Virginia enacted SB 384 this year, which directs the Joint Commission on Technology and Science to study the future development of an IVO framework (after earlier versions of the bill would have created an IVO licensing structure akin to this draft).
These provisions would require:
Access: Large frontier developers would have to provide IVOs with timely access to unredacted materials and other information reasonably necessary to conduct audits and assessments. Developers could impose reasonable security protocols and access limitations to protect trade secrets and confidential business information.
Audit and Assessment Content: IVO reports would have to describe the scope, time period covered, materials reviewed, methods used, and any limits on the assessment. They would also assess the developer’s compliance efforts; the adequacy of its frontier AI framework, and risk-monitoring; any “failure, deficiency, or material weakness;” and internal controls.
Ad Hoc Audits and Assessments: The Director of CASI could require additional audits when necessary to verify compliance, validate prior findings, or monitor significant changes in risk, including after a critical safety incident or substantial modification.
Supplemental Reports: IVOs could also need to submit supplemental reports when new information, unexpected model capabilities, or major model changes call into question an earlier finding that the developer was adequately managing catastrophic risks.
Additionally, the Bill would create a federal licensing and oversight system for the auditors themselves, making IVOs a potential new layer of AI governance.
Together, these proposals suggest that policymakers are increasingly looking to independent verification as a middle ground between company self-assessment and direct government review. The Great American AI Act would take that idea further by specifying who can serve as an independent verifier, what they must review, and how their findings may be shared with enforcement authorities.
Federal Preemption and the State AI Landscape
One of the draft’s most consequential provisions is its three-year preemption clause. The draft would prohibit states from establishing, continuing in effect, or enforcing any law or regulation that specifically regulates the development of an AI model (emphasis added). It would not preempt laws of general applicability, state authority granted under the draft, or laws governing post-deployment activities, including the implementation, distribution, offering, or use of AI systems, products, or services.
This approach differs significantly from the AI moratorium Congress considered last year as part of the One Big Beautiful Bill Act. That proposal would have broadly restricted states from enforcing laws or regulations that “limit,” “restrict,” or “regulate” AI models, AI systems, or automated decision systems, while allowing laws viewed as “facilitating” AI. It also evolved from a proposed ten-year pause to a five-year compromise before ultimately being removed from the package. By contrast, the Great American AI Act uses narrower language: it applies only to laws that specifically regulate AI model development and sunsets after three years.
However, as always, preemption would be complicated. Many state AI laws do not neatly separate development from deployment. They may impose obligations on developers and deployers, or require pre-deployment documentation or risk assessments for systems that are ultimately used in employment or other high-impact contexts.
As drafted, the draft’s preemptive effect would likely be greatest for state laws focused on frontier model development, such as California’s SB 53, New York’s RAISE Act, and Illinois SB 315. But the clause could also create uncertainty for other AI laws. For example, a chatbot law focused on how companies offer chatbots to the public would likely be easier to preserve under the post-deployment carveout. But if a law requires changes to how a model is trained, fine-tuned, tested, or designed before release, the line between deployment and development may be harder to draw.
The result is a preemption provision that is far narrower than last year’s proposed 10-year moratorium, but still raises significant interpretive questions. The answer to those questions is likely to have an impact on both developers and deployers, and would likely depend on how regulators and courts characterize specific laws and their requirements.
Other Key Provisions
Although the frontier model provisions are the most significant for most AI developers, the draft reaches well beyond AI safety: It also includes several provisions focused on the broader AI ecosystem, including workforce development and displacement. This section highlights three additional provisions worth watching: a study on government engagement with AI platforms, new federal AI testbeds and voluntary model testing, and disclosure requirements for AI-related mass layoffs.
Study on Protecting Free Speech: The draft would direct the Secretary of Commerce to study how the government encourages or pressures AI companies on content moderation, information prioritization, and output generation. The report must examine legal frameworks governing federal agency interactions with AI platforms and recommend legislation providing individuals with redress against unlawful government censorship, including transparency and oversight mechanisms. While this section is only a study requirement, it signals congressional interest in how the government communicates with AI companies about content moderation and model outputs. For AI companies, this study may lead to future regulation on transparency and government oversight.
AI Testbeds:The draft would direct the establishment of an AI testbed program for testing, evaluating, and assessing AI systems, including automated evaluations, security vulnerability assessments, and computational resource assessments. It would also establish a voluntary foundation model testing program for vendors of foundation models, AI virtual agents, and robots that incorporate foundation models. This provision parallels recent federal activity around voluntary AI testing and evaluation, including the White House’s June 2026 executive order establishing a voluntary framework for certain frontier AI developers to share models with the federal government before public release for national security and cybersecurity assessment.
Disclosures for AI Layoffs: The draft would require employers to provide 60 days’ advance notice when AI is a “substantial factor” in a mass layoff, including information on the AI involved, the estimated share of job losses attributable to AI, and pre-layoff upskilling or retraining efforts. A similar requirement was recently enacted in Connecticut’s SB 5, which requires employers issuing mass layoff notices to tell the state’s Labor Department whether the layoffs are related to the employer’s use of AI. This provision reflects growing interest in how AI may affect workforce displacement and whether existing worker-notice frameworks should account for AI-related job losses.
Conclusion
The Great American AI Act is still a discussion draft; whether it will advance in its current form remains uncertain. Federal AI legislation has faced a difficult path in Congress and the draft touches on issues, especially frontier model safety and preemption, that are likely to generate significant debate.
But the draft is still an important marker in the federal AI conversation. It shows that Congress is continuing to consider how to regulate frontier model development, how to structure independent oversight, and how federal rules should interact with the growing number of state AI laws. Even if this bill changes substantially, these questions are unlikely to go away.
In the meantime, states are likely to continue to test different approaches to AI regulation, including laws that regulate frontier model development, deployment, or both. That could make the draft’s preemption language especially important to watch. The next phase of AI policymaking could be defined not only by the rules new proposals would set, but also by the coalitions advancing them and the venues where they move forward: Congress, the Executive Branch, the states, or all three.
Five major federal chatbot proposals have been introduced: the CHAT Act (S.2714/H.R. 7218), GUARD Act (S.3062/HB 8623), SAFE BOTs Act (H.R. 6489), Youth AI Privacy Act (S. 4199), and the CHATBOT Act (S 4407). ↩︎
Rep. Obernolte (R-CA) previously co-chaired the bipartisan House Task Force on Artificial Intelligence, which released its final report in December 2024. Rep. Trahan (D-MA), a member of the House Energy and Commerce Committee’s Innovation, Data, and Commerce Subcommittee, has also been active on technology accountability issues. ↩︎
Privacy Becomes You, Bayou State: A Look at the Louisiana Data Privacy Act
Louisiana has become the 22nd U.S. state to enact a comprehensive consumer privacy law—and the third this year following Oklahoma and Alabama—after Governor Landry signed the Louisiana Data Privacy Act (LDPA) (SB 386) on May 29. Overall, this is a fairly standard state privacy law that follows the Washington Privacy Act framework apart from the law’s CCPA-style applicability thresholds. The law will go into effect on January 1, 2027. This blog post covers the LDPA’s scope, consumer rights, business obligations, and enforcement.
Scope
Applicability: Like other comprehensive privacy laws based on the Washington Privacy Act (WPA) framework, this law regulates controllers’ and processors’ collection and use of personal data.
Departing from the common WPA framework, this law’s applicability thresholds are modeled on those under the California Consumer Privacy Act (CCPA). The LDPA only applies to a person or entity doing business in Louisiana that either—
Has annual gross revenues exceeding $25 million;
Annually buys, “receives for the business’s commercial purposes,” sells, or shares for commercial purposes the “personal information” of at least 75,000 consumers, households, or devices; or
Derives 50% or more of its annual revenues from selling consumers’ “personal information.”
Not only do these thresholds reflect those under the CCPA, they also use the undefined term “personal information” (which is used in the CCPA) rather than the defined term “personal data” used throughout the LDPA. One unique aspect of these applicability thresholds is that prong (2) adds the criteria “receives for the business’s commercial purposes,” which is not present in the CCPA’s text although that law defines “commercial purpose” and uses the term in other contexts. (§ 1780.2(A).)
The LDPA includes broad entity- and data-level exemptions, including for—
State agencies or political subdivisions;
Financial institutions, affiliates, or data subject to the GLBA;
Covered entities, business associates, and protected health information governed by HIPAA;
Health records;
Information included in a limited data set maintained as required under 45 C.F.R. 164.514(e);
Nonprofits, including political organizations;
Institutions of higher education;
Information collected, used, or disclosed by a consumer reporting agency or furnisher to the extent regulated by and authorized under FCRA;
Personal data collected, processed, sold, or disclosed in compliance with the DPPA;
Personal data regulated under FERPA; and more. (§ 1780.2(B)-(C).)
Key Definitions: Personal data is defined consistently with other state laws as information that is linked or reasonably linkable to an identified or identifiable individual, and it does not include deidentified data or publicly available information. Sensitive data includes: personal data revealing racial or ethnic origin, religious beliefs, mental or physical health diagnosis, sexuality, or citizenship or immigration status; genetic or biometric data processed for uniquely identifying an individual; personal data collected from a known child (under 13); and precise geolocation data (within a radius of 1,750’). This definition is narrower than in many of the newer state laws, which often include other categories such as consumer health data, neural data, or status as a victim of a crime. (§ 1780.1.)
This law includes many of the key definitions associated with the Connecticut-model of state laws. For example: the definition of “biometric data” includes data generated from a photograph or video or audio recording if generated to identify a specific individual; “dark patterns” are defined and prohibited for obtaining consent; and “sale” is defined broadly to include exchanges of personal data for “other valuable consideration” apart from monetary consideration. (§ 1780.1.)
Consumer Rights
Consumers will have the standard rights to: confirm whether their personal data is being processed; access their personal data; correct inaccuracies in their personal data; have their personal data deleted; obtain a copy of their personal data in a portable format (if available in a digital format); and opt-out of the processing of their personal data for the purposes of targeted advertising, the sale of personal data, and profiling in furtherance of a decision that produces a legal or similarly significant effect concerning the consumer. Like the laws in Tennessee and Alabama, these consumer rights (including the opt-out right) do not apply to pseudonymous data if the controller is able to demonstrate that information necessary to identify the consumer is kept separately and subject to effective technical and organizational controls that prevent the controller from accessing the information. (§ 1780.3(A), 1780.4(O),)
Controllers must respond to consumer rights requests within 45 days, which can be extended an additional 45 days if necessary so long as the consumer is informed of the extension and the reason. If a controller declines to act on a consumer request, then it must inform the consumer of the decision, the justification, and how to appeal the decision. A controller is not required to comply with a rights request that it cannot authenticate, and that the authentication requirement extends to the consumer opt-outs as well. Some states, like Connecticut, provide that a controller does not need to authenticate an opt-out request but may deny an opt-out request if it has a good faith, reasonable and documented belief that the request is fraudulent. (§ 1780.3(B).)
In another departure from the Connecticut-style suite of state laws, the LDPA does not appear to require a controller to provide consumers with a mechanism to revoke previously given consent.
What about Agents and OOPS? A consumer will be able to designate another person to serve as the consumer’s authorized agent to opt-out of the processing of consumer’s personal data for targeted advertising or the sale of personal data. Although the law does not reference opt-out preference signals (OOPS) or universal opt-out mechanisms (UOOM), it does provide that a consumer can “designate an authorized agent using a technology, including . . . a global setting on an electronic device,” that allows the consumer to indicate the consumer’s intent to opt out of the processing for targeted advertising, for sale of personal data, or both.” Additionally, a “technology” described in the subsection may not “unfairly disadvantage another controller,” make use of a default setting (instead requiring “an affirmative, freely given, and unambiguous choice” by the consumer), and be consumer-friendly and easy to use. These are the common requirements for an OOPS under the state comprehensive privacy laws.
The use of a technologically-designated authorized agent by a consumer could be limited due to several exceptions under the law. A controller is not required to comply with an opt-out request from an authorized agent if: the authorized agent does not communicate the request in a clear and unambiguous manner; the controller cannot verify (with reasonable effort) that the consumer is a resident of Louisiana; the controller does not possess the ability to process the request; or the controller “does not process similar or identical requests the controller receives from consumers for the purpose of complying with similar or identical laws or regulations of another state.”(§ 1780.3(E)(5)-(6).)
Business Obligations
Consistent with most of the state privacy laws, controllers and processors are subject to an enumerated list of duties under the law—including transparency, data minimization, data security, non-retaliation, oversight of processors, data protection assessments, and children-specific protections—as well as a list of broad exceptions.
Transparency: Controllers must provide consumers with a “reasonably accessible and clear privacy notice” including information such as categories of personal data processed, processing purposes, how consumers can exercise their data rights, categories of personal data sold to third parties, and categories of third parties to whom data is sold. If the controller sells personal data, processes personal data for targeted advertising, sells sensitive data, or sells biometric data, there are additional notices that must be provided in the privacy notice (e.g., “NOTICE: We may sell your sensitive data”). (§ 1780.4(B).)
Data Minimization: The LDPA includes common procedural data minimization and purpose limitation restrictions. A controller must—
“[L]imit the collection of personal data to what is adequate, relevant, and reasonably necessary in relation to the purposes for which that personal data is processed, as disclosed to the consumer”
Obtain the consumer’s consent to “process personal data for a purpose that is neither reasonably necessary to nor compatible with the disclosed purpose for which the personal data is processed, as disclosed to the consumer”; and
Obtain the consumer’s consent to process the consumer’s sensitive data. (§ 1780.4(A).)
Data Security: A controller must establish, implement, and maintain reasonable administrative, technical, and physical data security practices that are appropriate to the volume and nature of the data. (§ 1780.4(A)(1)(b).)
Anti-discrimination and Non-retaliation: Controllers cannot process personal data in violation of state and federal laws that prohibit unlawful discrimination against consumers. Controllers also may not deny goods or services, charge different prices or rates for goods or service, or provide a different level of quality of goods or services to the consumer as retaliation for a consumer exercising any of their rights under the LDPA, subject to exceptions (e.g., if the data is necessary to provide a service or processed in connection with a bona fide loyalty program). (§ 1780.4(A).)
Processors: Processors must adhere to the instructions of a controller and assist the controller in complying with the controller’s duties or requirements under the law. Whether a person is acting as a controller or processor is a fact-based determination depending on context, but a processor remains a processor if they are adhering to a controller’s instructions with respect to a specific processing activity. There must be a valid contract in place between the controller and processor that meets statutory criteria (e.g., clear instructions for processing, deleting or returning personal data after the service is concluded). (§ 1780.4(D).)
Children’s Privacy: Consistent with most other state comprehensive privacy laws, the LDPA includes protections for children’s personal data and provisions that address COPPA compliance. “Sensitive data” includes personal data collected from a known child, and a controller must process the sensitive data of a known child in accordance with COPPA. Parents are able to exercise consumer rights on behalf of a child whose personal data is processed. Controllers and processors that comply with the verifiable parental consent requirements of COPPA are deemed to be in compliance with any requirement to obtain parental consent under the LDPA. In contrast to many of the newer state comprehensive privacy laws, the LDPA does not include opt-in rights for teenagers with respect to targeted advertising or the sale of personal data. (§§ 1780.1, 1780.2(E), 1780.3(A) & 1780.4(A).)
Data Protection Assessments: Controllers must conduct and document a data protection assessment for processing activities that present a heightened risk of harm to consumers, including processing personal data for targeted advertising, selling personal data, processing personal data for profiling that presents a reasonably foreseeable risk of substantial injury to consumers, and processing sensitive data. A controller must make a data protection assessment available to the Louisiana attorney general if requested in a civil investigative demand (although that requirement includes a cross-reference to a non-existent subsection of the law). (§ 1780.4(E).)
Exceptions: This bill includes a number of common exceptions, providing that nothing in the law shall be construed to limit a controller’s or processor’s ability to: comply with state, federal, or local laws or regulations; comply with regulatory inquiries or investigations; provide a specifically requested product or service; engage in public or peer-reviewed research in the public interest adhering to relevant safeguards; cooperating with law enforcement agencies; internal use of data for conducting research, effectuating a product recall, identifying and repairing technical errors, performing internal operations reasonably aligned with consumers’ expectations; and more. (§ 1780.4(G)-(I).)
Miscellaneous: This law includes one unique provision related to the sale of sensitive data. Section 1780.4(P) provides that “[a] person or entity described by R.S. 51:1780.2(A)(3) may not engage in the sale of personal data that is sensitive without receiving prior consent from the consumer,” and violation of that requirement subjects a person to a penalty under the law. The cross-reference is to the applicability threshold for a person or entity that does business in the state and that derives fifty percent or more of its annual revenues from selling consumers’ “personal information.” This is an ambiguous requirement. An entity meeting that threshold would already be under the requirement to obtain consent prior to processing sensitive data, which includes selling data, so it is not clear that this is an added responsibility, unless it is meant to apply more broadly. But there is no other language in the requirement suggesting that it would apply notwithstanding the law’s broad entity-level exemptions.
Enforcement
The Louisiana Attorney General will enforce the LDPA and violations will constitute unfair and deceptive trade practices under Louisiana’s Unfair Trade Practices and Consumer Protection Law. Notably, the private rights of action under the unfair trade practices law do not extend to violations of the LDPA. For the first six months of enforcement (January 1, 2027 to July 31, 2027), the attorney general must give persons notice of alleged violations and at least 30 days to cure those violations prior to initiating an investigation.
The attorney general is required to post online information regarding controllers’ and processors’ responsibilities and consumer rights under the law, and money received from enforcement actions will go towards funding the attorney general’s consumer protection efforts or promoting consumer protection and education.
Pictured: Louisiana receiving its star on the FPF “Privacy Patchwork” quilt.
Comparing Enacted App Store Accountability Acts
On May 28, 2026, the 5th Circuit granted a stay on the preliminary injunction blocking enforcement of Texas’s App Store Accountability Act (ASAA)—meaning the law is now in effect while litigation on the merits continues. In 2025, Utah, Texas, and Louisiana enacted App Store Accountability Acts (ASAAs) which impose novel and significant age assurance obligations on app store providers and app developers. These laws require account holder age verification and parental consent for minors at the app store level, with age band and consent data transmitted between the two parties via bespoke “age signals.” Texas’s law was the first scheduled to go into effect on January 1, 2026. Before the law took effect, however, two groups filed suit challenging its constitutionality on First Amendment grounds. In December 2025, a federal judge issued a preliminary injunction blocking the law, finding it more likely than not an unconstitutional content-based regulation. Attorney General Paxton quickly appealed this decision to the 5th Circuit seeking a stay on the injunction, which was subsequently granted last week.
The uncertain constitutional outlook of ASAAs puts compliance teams in a difficult position. Moving slowly risks liability if the laws survive legal challenge; moving quickly risks sunken compliance costs if ASAAs are ultimately struck down—costs that could have otherwise been diverted to other important trust and safety priorities. While Utah’s, Texas’s, and Louisiana’s laws impose broadly similar obligations on app store providers and developers, important distinctions will shape how companies ultimately achieve compliance. Amendments to Utah’s and Louisiana’s laws in 2026 further shape the developing obligations in each state. FPF created a comparison chart detailing the key terms, scope, and core obligations of these laws, including changes incorporated by 2026 amendments. The chart also includes an Appendix detailing current information about Developer APIs released publicly by app stores to aid developers as these requirements go into effect.
Scope: These laws apply broadly to apps publicly available to consumers for download in an app store with very narrow exceptions, such as for settings apps or essential device drivers, and no wholesale exemptions from application. Uniquely, Louisiana also applies to apps that are available for download from an app store onto connected devices.
Obligations: These laws require app store providers to implement age verification and obtain parental consent for minor account holders making purchases, downloads, in-app transactions, or significant account changes. Utah and Louisiana extend this requirement to a minor’s first-time access to many pre-loaded apps. Developers must request and receive age signals from app stores, use that data to manage minor access, and implement appropriate safety features. Utah and Louisiana add notable nuances for developers, such as allowing them to request that app stores block minor access to their apps entirely, or permitting reliance on a primary account holder’s age data for “family account apps.”
Enforcement: Private rights of action (PRAs) are a trending enforcement mechanism in ASAA models. While Utah relies solely on a PRA for enforcement, Texas’s law only allows a PRA through incorporated references to the Deceptive Trade Practices Act (DTPA). Louisiana is the only law that does not include a PRA.
The ASAA trend has continued into 2026 with Alabama enacting a new ASAA law back in February. Alabama’s law largely tracks with the ASAA models in Utah and Louisiana following their 2026 amendments. As ASAA legislation and litigation continues to develop, the 5th Circuit’s stay on injunction means that compliance teams must figure out how to navigate this growing thicket of app store age signals and online safety requirements in the meantime.
No Silver Bullet, But a Silver Lining? PETs and International Data Transfers
Is there a role for Privacy Enhancing Technologies (PETs) to play in the context of international data transfers? The answer to this question could be one of the keys to unlock trusted cross-border data flows at scale in the age of AI.
This was the topic explored in a session organized by the Future of Privacy Forum (FPF) during the Global CBPR Forum in Lima, Peru. Bringing together technical, regulatory, policy, and academic perspectives, the session provided an in-depth overview of initiatives centering on PETs in data transfer developments. It included a technical presentation of two of the most promising such technologies available – trusted execution environments and differential privacy, as well as a recent use case from a US-UK policy pilot and a regional perspective on PETs adoption in Latin America.
The session highlighted both the growing maturity of PETs deployments, as well as the structural challenges that continue to shape their adoption. A central takeaway was that PETs are increasingly being positioned as enabling tools for data use and collaboration — particularly in contexts where legal, regulatory, or trust constraints have historically limited data sharing. By illustrating a specific medical use case and their technical features, speakers demonstrated how PETs can support more responsible data ecosystems and trusted data sharing.
You can read the full blog on the Global CBPR Forum’s website here.
Career Choice in the AI Age: What Next for Privacy and Data Professionals?
When I was in college, privacy existed but the privacy profession did not. Some cynics might say that the reverse is true now, but the reality is more complex: even amidst mounting pressures on individual privacy, there are arguably more privacy protections enshrined into law around the world than ever before.
One point, though, is beyond debate: the privacy profession is changing. The rise of AI and AI governance work, under the broad umbrella of privacy and compliance work, raises questions about law and policy, how to be effective internally, how to take basic data governance and map it to AI governance, and how to create sustainable governance structures and processes. But it also raises fundamental – and familiar – questions about how we map our careers and what choices we make.
Many of the first privacy professionals started by doing other things and in many cases being appointed by their organization to handle the new privacy issues that arose. I started in government affairs and lobbying and bounced into privacy via trust and safety. Many privacy lawyers were simply lawyers in legal departments working on contracts or compliance or intellectual property and were “volunteered” to handle privacy.
Over time a new generation of privacy pros arose, intentionally choosing to work on these interesting new issues. Some loved learning the new technologies. Some were attracted to the idea of upholding the core and important values of privacy. Others liked the multidisciplinary nature of the work. It was a basket of “cool” legal and policy issues on the leading edge of technology.
Eventually the profession got more specialized, with engineers, project managers, non-lawyers and the like finding important roles. And over time, the uphill nature of the work caught up with some privacy pros (but not all), giving rise to burnout from constantly feeling like the organization resists the compliance work or concept of risk that privacy pros perceive.
AI is a major technological leap, raises new issues of law and policy and governance, and arrived so fast that few data or AI professionals were in a position to intentionally choose the work they do now. This has implications for how people might think about their careers in data, privacy, and AI.
One might characterize privacy professionals in four categories since AI swept in.
Adopting – They love the job and they’re making it their own. They think the tech is cool, they’re optimistic about what AI will be able to do. The law, policy, and governance challenges are interesting. This work gives them purpose and meaning.
Adapting — Realistic and pragmatic, figuring it out and making it work. The adapters might be struggling because they didn’t really wish for AI; they wouldn’t have chosen it. They’re in a privacy role and now they must adapt to this new technology and this new set of issues and this new pace of change. They try to make the most of it but now career questions arise: Is this the work they want to do? Purpose and meaning are in question.
Enduring — Hanging on, fighting the good fight, until something else feels more rewarding. A sense of obligation or duty, such as a need to support a family, supplants purpose and meaning.
Resisting – Resistance takes two broad forms. The change is too much, there’s no interest in the technology, fear and anxiety predominate, and change is resisted. Some people withdraw. These are the “quiet quitters” who don’t know what to do, think they are in a comfort zone and don’t want to leave it. Alternatively, people get angry and rebel, resist, push back and try to stop it or bring about meaningful change.
Where Are You Going?
Maybe you see yourself in one of these categories. Maybe you fit into a category not mentioned here (tell us about it!). For three of the four, career choices loom: Is this what I want to work on, work that chose me that I did not choose? What’s next?
Here are some helpful lenses or frameworks to think about the career choices and mapping we face:
Run toward, not away. It’s important to feel like you’re running towards something desirable rather than running away. That means you need to do the work to figure out what you want. That often connects to a deeper meaning or purpose.
Enlarge, not diminish. Another useful lens for looking at big decisions (career or otherwise) is an idea put forth years ago by James Hollis, a Washington DC area psychologist. He suggests asking this question: “Does this choice diminish me, or enlarge me?”2 It’s a powerful question. If working on AI makes you feel bigger in the organization, or it feels like growth to you or it makes you feel more influential more on top of things then that’s clearly a good choice – for you. If it makes you feel smaller, like you’re being pushed down in the organization, like you can’t keep up, like you’re overwhelmed, then maybe this isn’t a good choice and a good place to stay in your career.
Practice, practice, practice. What is the practice in your profession, the thing you seek to perfect? Surgeons seek to get better at surgery. Pianists practice to constantly get better. Professional athletes, lawyers, teachers, coaches, all seek to get better at what they do. That refinement of the craft, through daily practice, provides meaning and focus. What is it in your role that is the daily practice, the craft you seek to refine, the talent you seek to develop? If that question is hard to answer, it might be time for a change.
When in doubt, choose growth. So often what feels like languishing or burnout might simply be an absence of growth. Feeling like we are growing creates meaning and focus. Horizontal growth, the learning of new information, broadening one’s horizons, can be entertaining but insufficient. Vertical development – growing soft skills, ability to lead people and across teams, expanding impact in the org or profession – is more meaningful for people. Are you growing now? What would feel like growth to you?
What changes if you have a clear destination or desired outcome to move toward? What does it feel like to simply move and act and see where change takes you – or leaves you?
We hope we don’t have to simply endure. If we do, we might hope we’d have the courage to resist. In the end, we will all have to be adapters, but what could it feel like to intentionally choose? What would it feel like to find direction and purpose, to feel new growth and adopt one’s work, role, and the technological change underway?
It’s important to be clear about what’s happening with you, to identify which category above you’re in and find a lens through which to see your choices. If you’re just making it up as you go along, you might still find yourself in a place you don’t want to be. It’s normal human behavior to want to stay in our comfort zones, but how can a place where you merely endure and do not grow be a comfort?
Privacy and data professionals are multi-faceted and multi-talented people. They may be guided by core values like privacy, excited by technology, eager to launch new products, or simply do good work. The lenses set out above are not the only ones: you can start with why, think about what you long for, explore the Ikegai matrix, or any number of other ways to think about what’s next. But make a choice, rather than let events simply happen to you. Our careers are often set up as default opt-out, but we can choose to opt-in to the work that fulfills us. AI presents us with that choice.
Twelfth Night, Act II scene 5: “Be not afraid of greatness: some men are born great, some achieve greatness and some have greatness thrust upon them.” ↩︎
James Hollis, What Matters Most: Living a More Considered Life, p.13. “Ask yourself of every dilemma, every choice, every relationship, every commitment, or every failure to commit, “Does this choice diminish me, or enlarge me?” Do not ask this question if you are afraid to find the answer.” ↩︎
FPF Releases Practitioner Guides on Privacy Enhancing Technologies for Education Stakeholders
The Future of Privacy Forum (FPF) has released a suite of practitioner resources on Privacy Enhancing Technologies (PETs) for the education sector. Building on FPF’s 2025 landscape analysis of PETs adoption by State Education Agencies, the new resources move from landscape analysis to implementation considerations — providing audience-specific guidance for the three practitioner communities most responsible for handling student data: state education agencies and statewide longitudinal data systems, education researchers, and EdTech vendors.
FPF worked with AEM Corporation to develop the resources, which include three practitioner guides and a comparative reference chart covering seven PETs relevant to education data environments.
Addressing a Gap Between Awareness and Practice
FPF’s 2025 landscape analysis found that awareness of PETs among education practitioners remains limited, and that even practitioners who understand what PETs are often lack the use case guidance needed to match a specific technology to a specific workflow. The new guides are designed to close that gap. Each is written for its audience’s actual decision context —as a practical resource for the people who manage longitudinal data systems, design research studies, or build and test EdTech products using student data.
“State education agencies, researchers, and EdTech vendors all work with student data, but they face different risks, different analytical requirements, and different governance obligations,” said Jim Siegl, FPF Senior Fellow for Youth & Education Privacy. “These guides are designed to help each audience understand not just what PETs can do, but what each approach costs analytically — and how to make and document those tradeoffs responsibly.”
What the Guides Cover
Privacy Enhancing Technologies for State Education Agencies: A practical guide to privacy-preserving computation for state education data systems addresses the specific challenges of SEA and SLDS environments, where linked longitudinal records create both high analytical value and elevated re-identification risk. The guide explains how PETs can reduce how often student-level data must be copied, moved, or distributed to support analysis, and provides use case guidance for cross-agency computation, public reporting, and research partnerships. It also addresses a tension that is particularly acute in state education data: the student populations most at risk of re-identification — small districts, low-incidence disability categories, and rare demographic combinations — are often those for whom noise-based methods like Differential Privacy perform least well analytically.
Privacy Enhancing Technologies for Education Researchers: A practical guide to conducting education research with reduced data exposure addresses the analytical tradeoffs researchers need to understand before selecting a PET for a given study. Results produced under Differential Privacy carry an epsilon parameter that should be reported. Synthetic data findings require disclosure of generation methodology and fidelity validation. The guide frames PET selection as a methodological decision with implications for replication and publication, not just a data governance requirement.
Privacy Enhancing Technologies for EdTech Vendors: A practical guide to handling student data across product, testing, and analytics workflows addresses the range of vendor workflows — system testing, staff training, product analytics, and collaborative research with agencies — that involve student data and carry different PET requirements. The guide emphasizes that vendors operate under a dual obligation: to deliver useful analytics and product capabilities, and to handle student data in ways that honor the trust schools and agencies have placed in them. It includes guidance on transparency with agency partners when PET-protected outputs are shared, including disclosure of noise parameters and fidelity limitations.
The Comparison Chart
Accompanying the three guides is a comparative reference chart covering seven PETs — Differential Privacy, Synthetic Data, Federated Learning, Trusted Execution Environments, Secure Multi-Party Computation, Homomorphic Encryption, and Zero-Knowledge Proofs — across six dimensions: approach, benefits, limitations, example use case, data utility impact, and implementation complexity. The chart is designed as a standalone reference for practitioners who need to quickly orient to the PET landscape or compare options for a specific workflow, without reading all three guides in full.
Selecting the Right PET
A consistent theme across all three guides is that PET selection is a methodological decision, not a compliance checkbox. Each approach involves a tradeoff between privacy protection and analytical precision, and that tradeoff varies by method and by context. Differential Privacy introduces noise that grows more distorting as group sizes decrease. Synthetic data may misrepresent rare populations. Secure Multi-Party Computation and Trusted Execution Environments constrain which analyses can be run. Federated Learning reduces raw data exposure but can produce less accurate models when district data is heterogeneous.
The guides encourage practitioners to identify the acceptable level of analytical imprecision for their specific workflow before selecting a PET, to document that choice and its rationale, and to disclose relevant parameters — such as epsilon values for Differential Privacy or fidelity validation results for synthetic data — where outputs are shared or published. PETs work best when integrated into existing data governance frameworks rather than treated as standalone solutions.
FPF has actively contributed to shaping policy and practice around PETs through discussion papers, reports, stakeholder engagement, and its PETs Repository, launched in November 2024 as a centralized resource for practitioners seeking practical information about these technologies. The new practitioner guides extend that work by providing the audience-specific implementation guidance the landscape analysis identified as a critical gap.
SB 5 in Five: What to Know About Connecticut’s New AI Law
Connecticut’s SB 5 fits a lot of AI obligations into a small bill number. This week, Governor Lamont (D) signed the 39-section bill into law, creating new requirements across several fast-moving areas of AI policy, including companion chatbots, automated employment decision tools (AEDTs), social media, and provenance data. The law also includes provisions related to frontier AI whistleblower protections, AI-related layoff notices, and planning for a state AI regulatory sandbox, making it one of the broader state AI packages enacted this year. The law’s provisions phase in over time, with effective dates ranging from October 2026 to January 2028.
The law follows several years of debate in Connecticut over how to regulate AI, including last year’s SB 2, which cleared the Senate but ultimately fell apart after a veto threat from Governor Lamont over concerns that the bill could hamper innovation. SB 5 takes a different path. Rather than establishing a single comprehensive high-risk AI framework, it stitches together a set of more targeted obligations, alongside provisions focused on innovation, workforce development, and future study.
The result is a wide-ranging law that touches many of the AI issues currently moving through state legislatures. With so much packed into SB 5, here are five things to know about Connecticut’s new AI law.
Note: The Governor also signed SB 4, a broad privacy bill that establishes a data broker registry and accessible deletion mechanism, regulates data-driven pricing, updates the CTDPA, and regulates direct-to-consumer genetic testing, which is covered in FPF’s recent blog.
Companion chatbots get their Connecticut chapter
SB 5 gives companion chatbots their Connecticut chapter, adding the state to a growing list of jurisdictions (New York, California, Washington, Oregon, Nebraska, Idaho, Iowa, and Georgia) writing rules for chatbot systems that can sustain relationships with users. The law would impose baseline protections for all users, including safety protocols for suicidal ideation and clear non-human disclosures, while also introducing minor-specific safeguards such as parental tools to manage privacy and screen time, as well as limits on engagement-maximizing features. For those following this rapidly evolving area, FPF maintains a continuously updated chatbot legislation trackerthat monitors activity in this space.
Scope: SB 5 uses a detailed set of carveouts to narrow the systems covered, similar to Nebraska’s and Idaho’s laws. But Connecticut’s definition of “AI companion” is more targeted: it focuses on systems that provide adaptive, human-like responses and can sustain a relationship over time. One carveout is especially notable: the law excludes “narrow, task-specific” tools that provide outputs related to a discrete topic or function, so long as the tool’s primary function is not to discuss mental health. Similar narrow-task carveouts appear in other state chatbot laws, but Connecticut’s version appears narrower because the exclusion may not apply where the tool’s primary function is mental health-related.
Requirements for all users: SB 5 follows several other companion chatbot laws enacted this year in setting a familiar baseline: safety protocols, non-human disclosures, and safeguards to keep AI companions from presenting themselves as human. Operators would need to publicly post safety protocols using “evidence-based methods” to detect and “clinical best practices and expertise” to respond to user expressions indicating suicide, self-harm, or physical violence. Because “evidence-based methods” and “clinical best practices” are not defined, operators may face questions about what detection tools or clinical inputs are sufficient to meet that standard.
The law would also require clear non-human disclosures when an AI companion could reasonably lead a user to believe they are interacting with a human. Like Washington and Georgia, Connecticut applies a one-hour disclosure interval for minors and a three-hour interval for adults. Although the law distinguishes between users, the shorter minor-focused interval could become the practical default if operators choose to comply uniformly.
Minor-Specific Requirements: For minors, SB 5 moves beyond “tell users it is AI” and into the harder question of how companion chatbots are designed to interact and build relationships over time. Similar to Oregon’s law, Connecticut’s protections apply when an operator “knows or has reason to believe” that a user is under 18, a standard that may require operators to account for contextual signals, not just direct knowledge of age.
Similar to Washington’s chatbot law, SB 5 would require operators to prevent their chatbots from engaging in certain harmful conduct before providing an AI companion to a minor, including encouraging disordered eating or physical violence; romantic interactions; and manipulative techniques intended to extend engagement, such as encouraging isolation from family or friends or fostering inappropriate emotional dependence. Terms like “inappropriate emotional dependence” and “disordered eating” are not further defined, raising questions about how operators should distinguish benign interactions from those prohibited under the law. The law also includes a broader provision prohibiting operators from “optimizing user engagement in any manner that disregards” the minor-specific safeguards, which may extend the reach of the minor protections beyond listed outputs.
Finally, SB 5 also requires tools for parents and minors to manage screen time and account settings, a feature that appears in other state chatbot laws, including Idaho, Nebraska, and Georgia.
Enforcement: SB 5 would make violations an unfair or deceptive trade practice enforced by the Attorney General, keeping the law in the AG-enforcement lane rather than creating a private right of action. These requirements take effect January 1, 2027.
For employment AI, SB 5 asks for a heads-up, not an audit or assessment
Employment AI gets its turn in SB 5, with new transparency requirements for automated employment-related decision technologies (AEDTs) used to shape decisions about hiring, promotion, discipline, and discharge. The section draws from broader ADMT laws, including California’s CPPA ADMT regulations, the Colorado AI Act as originally enacted in 2024, as well as employment-specific laws such as New York City’s LL 144. But unlike other state frameworks, Connecticut does not require AEDTs to undergo bias audits or risk assessments. Instead, SB 5 focuses on disclosures and written notice to applicants and employees, similar to the revised Colorado ADM Act.
Scope: SB 5 is narrower than broader ADMT frameworks that apply across sectors such as housing and education. It covers AEDTs that are a “substantial factor” used to “make or materially influence” an employment-related decision. The law defines “substantial factor” as something that “meaningfully alters” the outcome, a narrower definition than the Colorado AI Act’s 2024 language covering systems that are “capable of altering” or “assist” in making a consequential decision.
Notice obligations: The notice obligations are allocated between actors in the AI value chain, similar to the current and previous version of the Colorado AI law. Beginning October 1, 2027, developers that market AEDTs for employment decisions would need to provide deployers with information about the tool. Deployers would then need to disclose to employees or applicants that an AEDT has been deployed, the purpose and nature of the decision, the tool’s trade name, the categories and sources of personal data used, how that data will be assessed, and contact information for the deployer. Developers and deployers do not need to disclose trade secrets, but must tell individuals when information is withheld on that basis.
Civil rights and enforcement: SB 5 may be notice-first, but it is not notice-only. The law also amends Connecticut’s human rights statute to clarify that using an AEDT is not a defense to discrimination claims, while allowing courts or the commission to consider evidence of anti-bias testing and related efforts when evaluating those claims. That consideration of anti-bias testing also builds on a related theme in last year’s amendments to the Connecticut Data Privacy Act, which included an exemption allowing controllers to process personal data for internal use to allow them to use data for bias testing. California and Illinois have similarly amended employment or human rights laws to address automated decision systems in the workplace. As a result, entities may not be required to conduct bias testing or assessments under Connecticut law, but are strongly encouraged to reduce their regulatory risk.
Violations would be treated as unfair or deceptive trade practices and enforced by the Attorney General, with a potential 60-day cure period through the end of 2027 and no private right of action. These requirements take effect October 1, 2026.
Social media and AI regulations increasingly become the dynamic duo in online safety
As the online safety landscape continues to evolve and other jurisdictions weigh pairing social media and chatbot regulations—Connecticut strikes first by incorporating a section on online safety obligations for social platforms into SB 5. Similar to laws enacted in California and New York, SB 5 restricts operators from providing minors under 18 access to a platform that “recommends, selects, or prioritizes for display…media items” shared by other users—also known as personalized recommender systems–unless certain requirements are met.
Age assurance and parental consent: As is commonly the case in social media frameworks, SB 5 requires that covered operators implement “commercially reasonable and technically feasible methods” to determine whether a user is an adult or a minor. In the case of a minor, a covered operator may not offer access to personalized recommender systems without first obtaining parental consent. Unlike California and New York, however, SB 5 does not authorize agency rulemaking to provide guidance on acceptable forms of age assurance under this law, potentially creating ambiguity for compliance teams.
Default safety features: The law also requires certain minor-specific default safety features seen in other recent frameworks, such as South Carolina’s Age Appropriate Design Code (AADC), including preventing unconnected users from viewing or contacting minor accounts and restricting minors from viewing “sensitive content.” Notably, SB 5 broadly defines “sensitive content” to include any material violative of platform community standards, “or any similar guidelines or standards” established by the covered operator. Lastly, in a novel move, covered operators would be required to limit minors’ access to personalized recommender systems to one hour per day by default, comparable to a recently enjoined obligation in Virginia. Only a minors’ parent can adjust the default time limit on personalized recommender system access through parental control mechanisms.
Parental controls: Covered operators must establish and provide parents or guardians access to prescribed controls for supervising the accounts of their children. These controls include providing parents the ability to prevent minors from receiving notifications outside of preset timeframes and limiting minor access to personalized recommender systems to specific times indicated by the parent. SB 5 would also require covered platforms to provide parents with a mechanism for setting the minor’s account to a protected mode that does not allow unconnected users to view published content of or exchange messages with minors—although it is unclear how this particular parental tool is supposed to be implemented alongside the seemingly identical default safety feature noted above.
Disclosures: SB 5 requires covered operators to provide two kinds of disclosures. First, minors must be provided a health warning from the Surgeon General concerning the potential harms of social media use. Secondly, covered operators must annually disclose to the state Attorney General’s office, in a publicly accessible format, information related to platform use, such as the total number of covered users for whom the covered operator obtained parental consent, enabled default settings, and the average amount of time covered users spent on the platform per day. SB 5’s reliance on disclosure obligations follows a growing trend of requiring various kinds of disclosures in online safety legislation to both individuals, like in Colorado’srecently enjoined social media warnings law, and to state entities for public accessibility, like in South Carolina’s AADC.
Enforcement: A covered operator’s violation of these requirements constitutes an unfair or deceptive trade practice under Connecticut consumer protection law, which includes a private right of action in addition to state enforcement authority. These requirements become effective on January 1, 2028.
AI provenance rules make their way east in SB 5
SB 5 picks up the provenance trend seen in western states–adding Connecticut to the growing list of states setting requirements for AI-generated content. SB 5 would require covered providers to include provenance data in content that is created or materially altered by a generative AI system. California spearheaded AI provenance data disclosure with the California AI Transparency Act, enacted in 2024 and amended in 2025. Other states with provenance data laws include Utah and Washington.
The provision is relatively targeted. It applies to covered providers that produce publicly accessible generative AI systems for personal use with more than 1 million monthly users. It also focuses on content that is created or “materially altered” by a generative AI system, while excluding minor modifications such as changes in color or resizing. That distinction helps to focus the law’s requirement on more meaningful generative AI edits, rather than changes unlikely to affect the substance of the content.
The law requires provenance data to be difficult to tamper with or remove. At the same time, covered providers are not required to include information relating to an identified or reasonably identifiable individual, trade secrets, or confidential or proprietary information.
These provenance requirements are narrower than SB 5’s provisions on chatbots or AEDTs, but still notable because they place Connecticut within a growing state-level push to make AI-generated and altered content easier to trace. Other provenance data bills are still pending in states like Arizona and New Jersey. These requirements take effect October 1, 2026.
Whistleblowers, layoff notices, and sandboxes get targeted treatment in SB 5
SB 5 also borrows from a few other state AI playbooks, like frontier AI protections and regulatory sandboxes. But in both cases, Connecticut takes a narrower path. Rather than creating a full frontier model governance framework or immediately launching a sandbox program, SB 5 focuses on employee whistleblower protections and planning for a potential future sandbox, as well as a targeted AI-related layoff notice requirement.
Frontier AI whistleblower protections: SB 5 borrows the language of frontier AI laws, but not the full architecture. Like California’s SB 53 and New York’s RAISE Act, it defines key terms such as “frontier developer,” “large frontier developer,” and “catastrophic risk.” But unlike broader frontier AI frameworks, SB 5 does not require developers to publish governance frameworks, issue transparency reports, or establish critical safety incident reporting mechanisms. Instead, it focuses on solely protecting employees who report certain serious AI-related risks.
The law would prohibit frontier developers from penalizing covered employees for protected whistleblower activity and bar retaliation against employees who report, with reasonable cause, conduct they believe poses a specific and substantial danger to public health or safety due to a catastrophic risk. Large frontier developers would also need to create an internal reporting process by January 1, 2027, allowing employees to anonymously report such risks, provide updates to reporting employees, share reports with directors quarterly, and notify employees of their rights. These requirements take effect October 1, 2026.
AI-related layoff notices: SB 5 also includes a workforce disclosure provision. Employers issuing plant-closing or mass layoff notices would need to disclose to the Labor Department whether the layoffs are related to the employer’s use of AI. These requirements take effect October 1, 2026.
Regulatory sandbox planning: SB 5 directs the Commissioner of Economic and Community Development to develop a plan for an AI regulatory sandbox program, joining a small but growing group of states (Utah, Texas, and Delaware) that have adopted AI sandbox frameworks. The program would allow approved applicants to test innovative AI systems under reduced regulatory requirements.
But here too, Connecticut starts with a blueprint. SB 5 requires planning for a potential sandbox, not the immediate launch of one, and asks the Commissioner to assess the feasibility of a reciprocal, multistate sandbox model. Recommendations are due by January 1, 2028.
Conclusion
SB 5 does not create one comprehensive AI framework. Instead, it reflects a broader trend in state AI policymaking of setting targeted obligations across several use cases, from companion chatbots and employment tools to provenance data and frontier AI employee protections. As states continue experimenting with issue-specific AI laws, Connecticut’s SB 5 offers another example of how significant AI regulation can emerge through issue-specific provisions. Additionally, as states continue to pursue substantive online safety frameworks for minors, whether other jurisdictions will pair social media regulation with chatbot safety requirements remains a trend to watch.
