H.R.8094 - AI Foundation Model Transparency Act of 2026
Issued by
U.S. Congress
- September 30, 2026 · Correction — Added the bill's thresholds, FTC rulemaking and enforcement, exemptions and committee referral, which the entry wrongly called unspecified (fact-check finding). (Cody Maxwell)
- October 1, 2026 · Correction — Rewrote the practical steps, newsletter hook, search description, and audience fields to match the entry's corrected content. (Cody Maxwell)
Bipartisan lawmakers introduced H.R.8094 on March 26, 2026. It would require large foundation model developers to disclose training data, model design, limitations, risks, and evaluation methods. The bill seeks public scrutiny through transparency without directly restricting model use or deployment.
Applies To
Overview
H.R.8094 would establish a federal disclosure regime specifically for foundation models, a category of large-scale AI systems trained on broad data and adaptable to a wide range of uses. Covered developers would have to publish standardized documentation on where training data came from, how the model is built, intended uses, known failure points, and evaluation methods used before release. The bill is framed as a transparency measure rather than an operational or safety mandate, meaning it does not prohibit specific model capabilities or require pre-market approval. The Federal Trade Commission would write the rules within a year of enactment, consulting NIST, Commerce, and OSTP. Violations would count as unfair or deceptive practices under the FTC Act. The bill was introduced with bipartisan support, signaling some cross-party appetite for AI disclosure requirements at the federal level. It was referred to the House Energy and Commerce Committee on 26 March 2026 and has not advanced further.
Key Requirements
- •A foundation model is trained on broad data, generally self-supervised, with at least 1 billion parameters.
- •Covered entities include developers with over 10 million monthly users or downloads, or models trained with more than 10^26 operations.
- •Models with high performance at tasks posing significant risks to security, consumer protection, civil rights, or public safety are also covered.
- •The FTC must write disclosure rules within 1 year, covering training data, intended uses and limitations, alignment efforts, benchmarks, and compute.
- •Violations count as unfair or deceptive practices under the FTC Act, with 14 days' notice before enforcement.
- •Fully open-source models are exempt, and small or new businesses get a three-month grace period.
What Your Organization Must Do
- →Inventory foundation models that could meet the bill's parameter, user, or compute thresholds.
- →Draft documentation now on training data sources, intended uses, limitations, and evaluations.
- →Track the bill at the House Energy and Commerce Committee, where it currently sits.
- →Check whether any models qualify as fully open source, which the bill would exempt.
- →Brief leadership that this is a pending disclosure bill, not a current legal duty.
- →Compare draft model documentation against EU AI Act disclosures to reuse existing work.
Playbook Guidance
Step-by-step implementation guidance for compliance teams.
Frequently Asked Questions
- Which AI models would be covered under H.R.8094?
- It covers foundation models with at least 1 billion parameters from developers above set thresholds. Those are over 10 million monthly users or downloads, training compute above 10^26 operations, or high performance at risky tasks. Fully open-source models are exempt.
- Does H.R.8094 restrict what foundation models can do or require pre-market approval?
- No. The bill is explicitly framed as a transparency measure, not an operational or safety mandate. It does not prohibit specific model capabilities, restrict deployment, or require regulatory approval before a model is released.
- What disclosures would foundation model developers be required to make under H.R.8094?
- Developers would disclose training data sources, data size and composition, data governance, intended purposes and limitations, alignment efforts, benchmarks, and compute used. The FTC would set the format in its rules.
- What are the penalties for non-compliance with H.R.8094?
- The FTC would enforce the rules, treating violations as unfair or deceptive practices under the FTC Act. It must give 14 days' notice before enforcement, and small or new businesses get a three-month grace period.
- How does H.R.8094 compare to the EU AI Act's transparency requirements for general-purpose AI models?
- Both regimes target large foundation or general-purpose AI models and require training data and capability disclosures. However, the EU AI Act also includes operational obligations and risk classifications, while H.R.8094 is limited to public disclosure without imposing usage restrictions or safety mandates.
- What should AI developers do now given that H.R.8094 has not yet been enacted?
- Inventory models that could meet the thresholds and draft documentation on training data, intended uses, limitations, and evaluations. The bill sits with the House Energy and Commerce Committee, so watch for markup.