FTC gets power to demand AI companies reveal training data and risks

What it does

This bill directs the Federal Trade Commission to create rules requiring AI companies that build large foundation models to publicly disclose information about their training data, how the models work, what they're tested for, and known risks—before selling them and throughout their use. Companies meeting size/capability thresholds (over 10 million users, trained on massive computing power, or capable of high-risk tasks) must submit detailed information to the FTC and publish consumer-friendly summaries on their websites; small businesses get a 3-month grace period and compliance help. The FTC enforces violations as unfair or deceptive practices.

Why we flagged it

The bill's core mechanism is a regulatory mandate requiring disclosure of AI training data, algorithms, and risk assessments to the FTC and public. It is fundamentally a transparency and accountability measure, not a subsidy, carve-out, or deregulation.

What the text implies

• Redaction provisions (subsection g) allow companies to withhold 'sensitive' information for cybersecurity or national security reasons, potentially creating loopholes for disclosure of genuinely harmful capabilities if companies claim security justification.
• The 'covered entity' definition relies on four alternative thresholds (high-risk capability, 10M users, 10M downloads, or 10^26 compute operations); a model could evade coverage by staying just below one threshold while remaining high-impact.
• Downstream models (built on top of existing models) need only link to the base model's disclosure and disclose their own 'significant changes'—the definition of 'significant' is left to FTC rulemaking, creating potential gaps in transparency for fine-tuned or adapted models.
• The bill requires FTC to update thresholds annually, but does not specify a process for public comment or congressional review of threshold changes, potentially allowing regulatory drift without legislative oversight.
• Open-source models are fully exempt, which may incentivize companies to release models as 'open-source' with minimal actual openness to avoid disclosure requirements.

Section numbers refer to the bill text the analysis read — linked under Primary records below.

Who it affects

Citizens gain transparency into how large AI systems are built and what risks they pose, enabling informed use and public accountability. The disclosure requirements apply only to high-impact models (over 10M users or trained on massive computing), and small businesses receive compliance assistance, reducing burden on smaller innovators.

Named in the bill

Federal Trade Commission (FTC), National Institute of Standards and Technology (NIST), Office of Science and Technology Policy (OSTP), Department of Commerce, AI foundation model developers, Small businesses and startups in AI sector

Where it stands

3 cosponsors: 2 Republicans, 1 Democrats.

• Mar 26, 2026 — Introduced · Congress.gov: “Introduced in House”
• Mar 26, 2026 — Referred to House Committee on Energy and Commerce · Congress.gov: “Referred to the House Committee on Energy and Commerce”

Dates and quoted wording are Congress.gov's action record; the timeline shows status changes, not every procedural step.

Money around this bill

3 lobbying clients named this bill on 4 disclosure filings across 1 quarter, Jun 2026 to Jun 2026. Those filings disclosed $1,955,000 in lobbying spend. A filing names 15 bills on average, so that figure is what each filing reported, not a share belonging to this bill.

More lobbying clients named this bill than 61% of bills with at least one filing.

Donald Beyer, the sponsor, reported $633,099 in PAC receipts in the 2026 cycle.

• Universal Music Group — $1,460,000 on 2 filings
• News Media Alliance — $475,000 on 1 filing
• Getty Images Inc. — $20,000 on 1 filing

Lobbying Disclosure Act filings through Jul 21, 2026. A filing shows who paid to lobby on a bill it names, not what changed.