The AI Accountability and Personal Data Protection Act (S. 2367) would create a consent tort that names the sale of data and generative AI training.
A bill imposes no duty on anyone today. This page reads each bill's introduced text and gives its status with a date, as a preview of the pending federal layer of AI training data licensing. The law in force sits on the parent page.
This page cites introduced text on govinfo.gov and bill-status records read on 2026-11-17, and its method is set out in how we source, date and check every figure.
Six Federal Bills at a Glance, and Where Each Stands
Six bills are listed below, each as introduced text with the status date its source gives, and none has passed either chamber.
The table compares the bills as of 2026-11-17, using the introduced text and the bill-status record, with the last data update for each bill in the status column; none of the rows is law.
| Bill | Citation and short title | Sponsor and introduced | What it would do | Status as of | Source |
|---|---|---|---|---|---|
| TRAIN Act (Senate) | S. 2455, 119th Congress | Welch, with Blackburn, Hawley, Schiff; 2025-07-24 | subpoena for training records of the requester's own works (new 17 U.S.C. 514) | referred to Senate Judiciary 2025-07-24; no later action recorded (bill-status data updated 2026-08-10) | govinfo.gov BILLS-119s2455is; congress.gov |
| TRAIN Act (House) | H.R. 7209, 119th Congress | Dean, with Moran; 2026-01-22; Lee added 2026-03-24, Casar 2026-07-21 | same text as S. 2455 after the short title | referred to House Judiciary 2026-01-22; no later action recorded (bill-status data updated 2026-08-20) | govinfo.gov BILLS-119hr7209ih |
| AI Accountability and Personal Data Protection Act | S. 2367, 119th Congress | Hawley, with Blumenthal; 2025-07-21; Welch added 2025-07-24 | federal tort for using or selling covered data, including generative AI training, without express prior consent | referred to Senate Judiciary 2025-07-21; no later action recorded (bill-status data updated 2026-08-10) | govinfo.gov BILLS-119s2367is; congress.gov |
| AI Foundation Model Transparency Act of 2026 | H.R. 8094, 119th Congress | Beyer, with Lawler, Jacobs; 2026-03-26 | FTC rules requiring covered entities to submit and publish foundation model training data information | referred to House Energy and Commerce 2026-03-26; no later action recorded (bill-status data updated 2026-04-09) | govinfo.gov BILLS-119hr8094ih |
| NO FAKES Act of 2026 | S. 4591, 119th Congress | Coons, 13 original cosponsors; 2026-05-20 | federal voice and likeness right, written licenses capped at 10 years while living | reported with a substitute 2026-06-24, Senate Calendar No. 446; no later action recorded (bill-status data updated 2026-08-10, read 2026-11-17) | govinfo.gov BILLS-119s4591rs; congress.gov |
| Generative AI Copyright Disclosure Act of 2024 | H.R. 7913, 118th Congress | Schiff; 2024-04-09 | notice to the Register of Copyrights of copyrighted works in a training dataset, 30 days before release | referred to House Judiciary 2024-04-09; no other action recorded (bill-status data updated 2024-06-11) | govinfo.gov BILLS-118hr7913ih |
The table counts the TRAIN Act twice, as Senate and House text, and includes one bill of the 118th Congress. "Referred" means sent to a committee, "reported" means the committee sent the bill to the full chamber, and "calendar" means the bill waits for a floor vote.
The TRAIN Act: A Subpoena for Training Records
The TRAIN Act (Transparency and Responsibility for Artificial Intelligence Networks Act) would let a copyright owner ask a US district court clerk for a subpoena that orders a developer to disclose copies of, or records identifying, the owner's works used to train a generative AI model.
The Senate bill, S. 2455, was introduced on 2025-07-24 by Senator Welch with Senators Blackburn, Hawley and Schiff and was referred to the Judiciary Committee.
The House text, H.R. 7209, was introduced on 2026-01-22 by Representative Dean with Representative Moran and is identical to S. 2455 after the short title, by text comparison on 2026-11-17. The bill was first introduced in the 118th Congress as S. 5379 on 2024-11-21, and the text of that version was not compared.
The text would add a new section 514 to title 17. The requester states a good-faith belief and files a sworn declaration, the request covers only the requester's own works, and the clerk issues the subpoena if the paperwork is in proper form.
A developer must then disclose the records expeditiously, and the recipient of the records may not share them without authorization.
If a developer does not comply, the bill creates a rebuttable presumption that the developer made copies of the works, and a bad-faith request is open to sanctions under Rule 11(c) of the Federal Rules of Civil Procedure.
The bill takes effect on enactment. It defines a developer as a person that designs, codes, produces, owns or substantially modifies a generative AI model and curates its training dataset or trains it, and it names third-party training dataset curators inside that definition.
A seller reading the text would note that the bill reaches the developer's records, so the record keeping and disclosure clauses in AI data licensing agreements are where such a subpoena would land.
The text does not say whether a data licensor that builds no model is a developer, and this page does not claim it.
S. 2367: A Federal Consent Tort That Names Sale and Training
The AI Accountability and Personal Data Protection Act (S. 2367) would make any person liable to an individual who appropriates, uses, collects, processes, sells or otherwise exploits that individual's covered data without express, prior consent, and its definitions name the training of a generative AI system.
Senator Hawley introduced the bill on 2025-07-21 with Senator Blumenthal, Senator Welch became a cosponsor on 2025-07-24, and it was referred to the Judiciary Committee. Covered data would include information tied to an individual, derived or inferred data, and copyrighted works the individual generated, with unique identifiers, geolocation, biometric and behavioral data named.
The bill would give the individual a private action in federal or state court, against the actor and against anyone who aided and abetted. The private action in the text introduced on 2025-07-21 provides:
- compensatory damages of the greater of actual damages, treble profits or USD 1,000;
- punitive damages;
- injunctive relief;
- attorney's fees and costs;
- liability for those who aided and abetted the exploitation.
Consent would be a defense, but it would be invalid if obtained by coercion or deception, or as a condition of using a product beyond what is reasonably necessary.
Where data goes to or through a third party, each third party must be specifically and clearly disclosed when consent is sought, and the disclosure must be presented distinctly and separately from any privacy policy, so a hyperlink to a policy is not enough.
A seller reading the text would conclude that consent collected under a general privacy policy does not meet the third-party rule, which makes the question for a customer-data or user-content license whether a separate, named consent exists, as in a separate customer consent and confidentiality review.
Predispute arbitration agreements and joint-action waivers would not be enforceable for claims under the bill. Section 4 would not preempt state law and sets a minimum standard. The introduced text contains only four sections and states no effective date, so this page states none.
Disclosure and Likeness Proposals: H.R. 8094, H.R. 7913 and S. 4591
Three other bills would add public or regulator-facing duties rather than a records subpoena: a likeness right, an FTC transparency rulemaking and a copyright notice.
The NO FAKES Act of 2026 (S. 4591) was introduced on 2026-05-20 by Senator Coons, ordered reported with a substitute on 2026-06-18 and reported on 2026-06-24 as Calendar No. 446, with no later action recorded on 2026-11-17. It would cap written licenses for digital replicas at 10 years while the person lives, and the detail sits on the page for voice and likeness rights in AI training data.
The AI Foundation Model Transparency Act of 2026 (H.R. 8094) was introduced on 2026-03-26 by Representative Beyer with Representatives Lawler and Jacobs and referred to the Energy and Commerce Committee.
It would have the FTC issue regulations within 1 year of enactment, under which covered entities submit information to the FTC and publish a sufficiently detailed summary of training data sources, how the data was collected, and its size and composition.
This page does not state who a covered entity would be, because that definition was not read.
The Generative AI Copyright Disclosure Act of 2024 (H.R. 7913) was introduced on 2024-04-09 by Representative Schiff in the 118th Congress and referred to the Judiciary Committee.
A person who creates or significantly alters a training dataset used in a generative AI system made available to consumers would file a notice with the Register of Copyrights 30 days before release, with a summary of the copyrighted works used.
As introduced on 2024-04-09, the bill sets a civil penalty of at least USD 5,000, a public database and an effective date 180 days after enactment.
A 119th Congress version was not confirmed, and this page does not say the bill was reintroduced or ended. A bill with a similar name, H.R. 3775, the AI Training Extension Act of 2025, concerns workforce training in the executive branch, not training data.
The disclosure rules in force today, for general-purpose model providers in the EU and for developers in California, are compared under training data transparency rules.
Benefits, Risks, Downsides and Alternatives for a Seller Watching These Bills, Including Not Selling
These bills create no duty today, so a seller can read them as a preview of questions buyers and rights holders may ask, but none should be treated as a deadline, and not licensing remains a valid choice.
Benefits
- None of the six bills imposes a compliance duty as of 2026-11-17.
- The texts show the questions to answer now: what is in the dataset, who consented to what, and who may see the records.
- Rules already in force require several of the same steps, as the border section below shows.
Risks
- S. 2367 as introduced would reach a sale of covered data and would void arbitration waivers for claims under it.
- The TRAIN Act subpoena would reach developer records, with a presumption of copying if the developer does not comply.
- A bill can change in committee, as the NO FAKES substitute of 2026-06-18 shows.
Downsides
- S. 2367 states no effective date and no bill has a passage date, so there is nothing to plan against.
- Drafting contract clauses around texts that may change costs time.
- A separate, named consent takes longer to collect than a privacy policy notice.
Alternatives
- License only data with documented, named consent, or data with no personal data in it.
- Add a change-in-law clause to a license, an editorial drafting choice rather than a requirement.
- Wait for committee action and read the final text.
- Not selling or licensing at all, among the alternatives to selling data to AI labs.
Beyond This Page: The US Law in Force Around These Bills
The bills sit on top of a federal layer that is already law. The full federal and state layer is covered under United States AI training data law, and the pages below take each part of it:
- AI training data law by jurisdiction: the same questions outside the US.
- Fair use and AI training: the fair use law the TRAIN Act would sit beside.
- US Copyright Office report on generative AI training: the agency's analysis of training.
- California AB 2013 training data transparency: the disclosure law already in force in California.
- EU AI Act Article 53: the EU training data summary duty.
- AI copyright lawsuits over training data: the court cases that move faster than these bills.
- Privacy law and AI training data compared: privacy rules compared by jurisdiction.
How This Page Is Sourced and Dated
Every bill here cites its introduced text on govinfo.gov and the status record on congress.gov or the bill-status data published at govinfo.gov, with the date it was read; congress.gov itself did not answer on 2026-11-17. This page is general information about pending bills, not legal advice for any specific deal.
Frequently Asked Questions
Is the TRAIN Act law?
No: the TRAIN Act is a bill (S. 2455 and H.R. 7209) that was introduced and referred to the Judiciary Committees, with no later action recorded. The status was read on 2026-11-17 from bill-status data last updated 2026-08-10 for the Senate bill and 2026-08-20 for the House bill.
Is there a federal law on AI training data?
No single federal statute governs AI training data as of 2026-11-17. Copyright, sector statutes and the FTC Act apply, and the bills on this page are proposals.
What does the AI Accountability and Personal Data Protection Act do?
Nothing yet, because S. 2367 is a bill referred to the Senate Judiciary Committee on 2025-07-21. If enacted as introduced, it would make a person liable for using or selling covered data, including for generative AI training, without express, prior consent.
Does a pending bill change a license signed today?
No: a bill binds no one until it is enacted. A license is read under the law in force when it is signed plus its own change-in-law terms.