Inside HR 8094: The Bipartisan US Bill That Would Force AI Labs to Open Their Training Books

The Bill That Tests Washington’s AI Patience

On 26 March 2026, a bipartisan trio introduced HR 8094 — the AI Foundation Model Transparency Act of 2026. The authors — Don Beyer (D-VA), Mike Lawler (R-NY), and Sara Jacobs (D-CA) — pitched the bill as a minimum-viable transparency framework for the handful of AI labs whose decisions now shape the information ecosystem for hundreds of millions of Americans. The pitch, per Beyer’s office, is less about restricting AI and more about requiring frontier labs to “show their work.”

The bill does not create a licensing regime. It does not set safety thresholds. It assigns the Federal Trade Commission — in consultation with NIST, the Department of Commerce, and the OSTP — to define what information covered foundation models must file with the FTC and what must be disclosed publicly. That design choice is deliberate: rather than legislating technical specifics that will age poorly, Congress delegates to agencies with existing standards-writing muscle.

What the Bill Actually Requires

Per the Benton Institute’s summary, the FTC’s disclosure regime would cover three buckets:

Training data. A sufficiently detailed summary of the data used to train the model — enough for regulators, courts, and journalists to evaluate provenance, copyright exposure, and bias dynamics, without forcing labs to publish raw corpora.

Training methodology. How the model was trained — architectural choices, fine-tuning stages, RLHF and safety tuning. Again, a summary adequate for outside review, not a full recipe.

User data collection. Whether and how user data is collected during use — a direct answer to the growing question of whether conversations with chatbots become training data for the next model.

The Alston & Bird Privacy Blog notes that the bill is carefully scoped to avoid the chilling effects that stricter early drafts raised: fully open-source models are exempt, enforcement sits with the FTC (not a new AI agency), and the summary-not-full-disclosure approach leaves room for proprietary protections.

Who Counts as a “Covered Entity”

HR 8094 applies only to foundation models that cross one of three thresholds, as detailed in the Congress.gov bill text:

1. Risk threshold — the model poses significant risks to security, civil rights, or public health.
2. Scale threshold — more than 10 million monthly users or downloads.
3. Compute threshold — trained using more than 10²⁶ computational operations (consistent with the threshold in President Biden’s 2023 AI Executive Order and widely referenced in subsequent US AI policy documents).

Fully open-source models are exempt. This carve-out matters politically — it’s what wins support from the open-source community and keeps the bill from looking like a moat for incumbent closed-model labs.

Why “Transparency-Only” Is Its Selling Point

The US has spent three years cycling through AI regulatory proposals. Bills on safety evaluations, pre-deployment licensing, and compute reporting have stalled. HR 8094’s design lesson, per the BGov analysis, is that transparency obligations may be the lowest-ambition framework that can still pass — narrow enough for Republicans wary of European-style regulation, meaningful enough for Democrats who want accountability.

The bill’s endorsement list, per Beyer’s office, spans industry, labor, and civil society — an unusual coalition in AI policy. That coalition is what makes HR 8094 a realistic vehicle rather than a symbolic gesture.

Current Status and What to Watch

As of April 2026, the bill has been referred to the House Committee on Energy and Commerce, which holds jurisdiction over FTC authorities. GovTrack rates it as a live bill with a non-trivial path to markup. Three variables will decide its fate:

• Committee calendar. If the Energy and Commerce Committee holds a markup before the August 2026 recess, the bill has a real shot at floor consideration.
• Senate companion. A bipartisan Senate version is reportedly in drafting; without one, the bill risks dying in the House.
• Administration signal. The bill gives the FTC new regulatory authority; the current administration’s comfort with that authority is a political variable.

For AI labs — OpenAI, Anthropic, Google DeepMind, Meta, Mistral, xAI — the near-term action is to begin building the internal documentation pipelines that Article 50 in Europe already requires, because if HR 8094 passes, the disclosure universe expands from voluntary model cards to legally binding FTC filings. The labs that have already built robust model cards have the shortest distance to travel.

Sources & Further Reading

• HR 8094 — AI Foundation Model Transparency Act of 2026 — Congress.gov
• Beyer, Lawler, Jacobs Introduce Bipartisan Legislation — Rep. Don Beyer
• Lawmakers Introduce Bipartisan Legislation to Promote AI Foundation Model Transparency — Benton Institute
• “Show Your Work, AI”: Congress Pushes for AI Model Transparency — Alston & Bird
• AI Transparency Requirements Emerge as Congress Crafts Framework — Bloomberg Government
• AI Foundation Model Transparency Act of 2026 — GovTrack