AI Systems Transparency Act
- Sponsor
- Sen. Coons, Christopher A. [D-DE]
- Committees
- Commerce, Science, and Transportation Committee (primary)
- Last action
- Sep 23, 2026
Bottom line
This bill establishes a federal framework for mandatory transparency and disclosure requirements for significant AI models and systems, aiming to enhance public trust and safety, particularly for vulnerable users.
What it actually does
The AI Systems Transparency Act mandates that developers of 'covered models' and 'covered deployers' of 'covered systems' publicly disclose comprehensive information about their artificial intelligence systems. This includes details on safety protocols for vulnerable users (especially minors), privacy measures, system capabilities and limitations, and evaluation practices. The Federal Trade Commission (FTC), in consultation with the National Institute of Standards and Technology (NIST), is tasked with establishing regulations for these disclosures and enforcing compliance.
Proponents argue
Supporters argue that increased transparency in AI systems is crucial for public safety, accountability, and fostering trust in rapidly evolving technology. They contend that mandatory disclosures, especially regarding safeguards for minors and vulnerable users, will allow independent researchers and the public to better understand, evaluate, and mitigate potential harms, promoting responsible AI development and deployment.
Opponents contend
Opponents might argue that the disclosure requirements could stifle innovation by imposing significant compliance burdens on AI developers, particularly smaller entities. Concerns could also be raised about the potential for trade secrets to be inadvertently disclosed, or for information to be misused by malicious actors, despite provisions in the bill designed to prevent such disclosures.
The bill is moderately detailed but clearly structured, making it accessible for an informed reader to understand its core provisions within a reasonable timeframe.
Section 2
Establishment of Key Definitions for AI Systems and Users
This section defines critical terms that determine the bill's scope, such as 'covered deployer' and 'covered model,' which are AI entities or systems meeting specific user thresholds (e.g., 1 million monthly users in the U.S., or 100,000 if minors can create accounts). It also defines 'generative artificial intelligence model' based on its characteristics and introduces the concept of 'vulnerable user,' encompassing minors, seniors, individuals with disabilities, and those experiencing mental health crises.
Supporters argue
Proponents argue that clear definitions are essential for effective regulation, ensuring that the bill targets the most impactful AI systems while providing specific protections for vulnerable populations who may be disproportionately affected by AI harms.
Critics contend
Opponents might argue that the user thresholds could be arbitrary or create a 'regulatory cliff' for growing companies, and that the definitions of AI systems may become quickly outdated given the rapid pace of technological advancement, potentially leading to over- or under-regulation.
Tradeoffs
The tension lies in crafting definitions that are broad enough to cover emerging AI technologies and protect users, yet specific enough to provide regulatory clarity and avoid stifling innovation for smaller entities.
Section 3
FTC Guidance on AI Model Versioning and Designation of High-Risk Models
This section directs the Federal Trade Commission (FTC), in consultation with the National Institute of Standards and Technology (NIST), to issue guidance on when a generative AI model is considered a 'distinct' version. It also requires the FTC to establish criteria for designating AI models or deployments as 'covered models' or 'covered systems' if they pose a significant risk to the safety of minors, vulnerable users, or the public. These designations will be accompanied by public written findings and a notice period for developers.
Section 4(b)
Mandatory Initial Public Disclosure Requirements for AI Model Developers
This is a core provision requiring developers of covered AI models to make extensive information publicly available. This includes details on consumer transparency, safety protocols for vulnerable users (e.g., privacy measures, filtering of child sexual abuse material, reporting procedures for harmful outputs, and research on mental health impacts). It also mandates system overviews (intended uses, capabilities, limitations), post-training enhancement methodologies, and safety/security protocols, including risk profile evaluations (e.g., chemical, biological, radiological, nuclear, cyber risks) and performance on public benchmarks.
Section 4(c)
Mandatory Ongoing Public Disclosure Requirements for AI Model Developers
Beyond initial disclosures, this provision requires developers to continuously update the public on any 'significant change' to the previously disclosed information, as defined by the FTC and NIST. It also mandates ongoing reporting of usage information, including data dictionaries for significant datasets, common violations of acceptable use policies (especially those pertaining to minor safety), how often prohibited prompts are refused, and attempts by users to generate illegal content like child sexual abuse material or non-consensual intimate imagery. Information on system capabilities for developing cyber, chemical, biological, radiological, and nuclear threats must also be disclosed.
Section 4(d)
Special Rules for Protecting Sensitive Information in Disclosures
This section provides specific exemptions from disclosure requirements for information that, as determined by the FTC, would compromise cybersecurity or safety practices of the developer, or could materially assist in the development, acquisition, or deployment of chemical, biological, radiological, nuclear, or offensive cyber weapons or capabilities. However, developers are still required to disclose whether an evaluation in a specified risk domain was conducted and a general characterization of the results, even if specific details are withheld.
Sections 4(e) and 5
Disclosure Mandates for Substantial Modifiers and Covered Deployers
Section 4(e) requires the FTC, in consultation with NIST, to promulgate regulations for 'substantial modifiers'—entities that significantly alter a covered model through post-training enhancements—to make appropriate disclosures. Similarly, Section 5 mandates the FTC to establish disclosure requirements for 'covered deployers' of 'covered systems,' consistent with the developer requirements but tailored to their specific role in deploying AI. This ensures that transparency obligations extend beyond the original developer to other key actors in the AI ecosystem.
Section 6
Enforcement by the Federal Trade Commission
This section designates violations of the Act or its regulations as 'unfair or deceptive acts or practices' under the Federal Trade Commission Act. This grants the FTC the full scope of its existing enforcement powers, including the ability to conduct investigations, issue cease-and-desist orders, impose civil penalties, and seek consumer redress. The FTC is also mandated to promulgate necessary rules to carry out the Act within one year of enactment.
The specific user thresholds that define 'covered deployer' (1,000,000 unique monthly users in the US, or 100,000 if minors can create accounts) and 'covered model' (deployed in covered systems meeting these same thresholds).
Section 2(6) and 2(7)
Why it matters:This is standard legislative practice for defining the scope of a bill's applicability, but its placement can obscure the precise trigger for regulation.
Case for: Defining clear user thresholds provides a measurable standard for determining which AI systems and entities are subject to regulation, focusing oversight on those with the broadest public impact.
Case against: Critics might argue that these thresholds could create a 'regulatory cliff,' where a company's rapid growth suddenly triggers significant compliance costs, potentially disincentivizing innovation or growth just below the threshold.
Estimated impact: Determines which AI developers and deployers, particularly those with large user bases or those serving minors, will be subject to the bill's extensive disclosure requirements.
The requirement for ongoing public disclosures of 'any significant change (as defined by the Commission in coordination with the Director of the National Institute of Standards and Technology, including any substantial modification to, or observed differences from, the most recent prior public disclosure under this section, and not including any immaterial variation for which a disclosure requirement would overly burden platform operations)' to previously reported information.
Section 4(c)(2)(A)
Why it matters:Delegating the definition of 'significant change' to the FTC and NIST allows for flexibility as AI technology evolves, which is standard legislative practice for complex technical areas. However, it also means the precise burden on industry is not fully clear in the bill text.
Case for: This flexibility allows regulators to adapt the definition of 'significant change' as AI technology evolves, ensuring that disclosures remain relevant and do not become overly burdensome for minor, inconsequential updates.
Case against: Critics might argue that leaving such a critical definition to agency rulemaking creates regulatory uncertainty for AI developers, making it difficult to plan for compliance costs and potentially leading to inconsistent application.
Estimated impact: Will determine the ongoing compliance burden for AI developers and the timeliness of information available to the public regarding updates to AI models.
The requirement for developers to disclose 'data dictionaries or equivalent documentation that explain significant datasets, as defined by the Commission, in the possession of a developer related to the covered model, including datasets that the developer maintains relating to the users of the relevant covered platforms, usage of the covered platforms, content generated on covered platforms, and actions taken to enforce the terms of service of covered platforms.'
Section 4(c)(2)(B)(i)
Why it matters:This provision is technically detailed and specific, making its placement within a list of other disclosure requirements typical. However, its profound impact on data governance and privacy makes it a 'buried treasure.'
Case for: Requiring data dictionaries provides unprecedented transparency into the data used to train and operate AI models, as well as user data collected by platforms. This is crucial for auditing for bias, understanding privacy implications, and ensuring responsible data handling.
Case against: Critics might argue that this requirement could expose highly sensitive proprietary data practices or create significant administrative burdens for documenting vast and complex datasets, potentially revealing competitive secrets.
Estimated impact: Significantly increases transparency into AI training data and platform data collection practices, with potential implications for privacy, bias detection, and intellectual property.