Open Source AI Bill: What H.R. 10152 Does

The Open Source AI Bill, H.R. 10152, would direct the US Department of Commerce to promote American open AI models and publicly assess risks from models tied to foreign adversaries. A House Energy and Commerce subcommittee advanced the proposal on September 1, 2026, but it is not law and still faces the full committee, House and Senate.

Key takeaways

H.R. 10152 is formally titled the Open-Source AI Leadership Act.
It would make Commerce identify adoption barriers and coordinate support for qualifying American open models.
It would require public assessments comparing foreign-adversary models with American alternatives.
The bill does not authorise Commerce to ban open models in interstate or foreign commerce.
Its central trade-off is faster adoption versus a government-led risk and nationality test.

The bill arrived as Chinese developers gained influence through downloadable open-weight models while many leading US systems remained accessible mainly through company-controlled services. That competition gives Congress a commercial reason to support domestic alternatives and a national-security reason to scrutinise foreign ones.

Everyone else is reporting an AI race with China; we are explaining the policy mechanism. H.R. 10152 does not fund a new national model or force companies to publish code. It uses Commerce Department coordination, barrier removal, agreements, monitoring and public comparison reports to steer adoption.

What happened to the Open Source AI Bill?

Representative Gabe Evans of Colorado introduced H.R. 10152 on August 27, 2026. The bill was referred to the House Committee on Energy and Commerce. On September 1, the Subcommittee on Commerce, Manufacturing, and Trade approved it for consideration by the full committee.

A subcommittee vote is an early legislative step, not final approval. The full committee can amend, advance or stop the proposal. If it reaches the House floor and passes, the Senate would need to approve compatible text before a president could sign it.

Bill fact	Verified detail
Official title	Open-Source AI Leadership Act
Bill number	H.R. 10152
Sponsor	Rep. Gabe Evans, Colorado
Introduced	August 27, 2026
Latest verified step	Passed House Energy and Commerce subcommittee on September 1
Lead agency if enacted	US Department of Commerce

The proposal has cleared only an early committee stage.

What would the Commerce Department do?

The bill tells the Commerce secretary to support the adoption and use of “qualified open models” in domestic and international commerce. The available text gives the department several tools rather than one fixed programme.

Designate a single contact to coordinate with government, companies, startups and other stakeholders.
Review existing Commerce programmes and use them, where appropriate, to support qualified open models.
Identify barriers that slow adoption and take appropriate action to address them.
Enter agreements with private entities, agencies, states and qualifying foreign partners.
Develop policies, recommendations and monitoring frameworks for open-model adoption.

This structure matters because it relies heavily on administrative judgement. “Appropriate” support could range from guidance and convening to changes in programme priorities. The bill’s effect would therefore depend on implementation, staffing and how Commerce defines measurable adoption barriers.

Which models would qualify?

The proposal distinguishes models developed and made available by US persons from models associated with covered nations or controlled entities. That nationality screen is central to the policy: the government would promote one category while studying and publicising risks in another.

“Open source” is often used loosely in AI. A model may release downloadable weights without releasing training data, full source code or the recipe needed to reproduce it. H.R. 10152 uses statutory definitions, so its practical reach will depend on the bill’s exact qualification tests rather than marketing labels.

The bill combines domestic promotion with foreign-model risk reporting.

What risks would the bill examine?

The sponsor’s summary says Commerce would assess risks involving personal and proprietary information, supply-chain security, model accuracy and censorship, safeguards against misuse, and chemical, biological, radiological, nuclear or other national-security capabilities. It would also compare foreign models with American alternatives on adoption, cost, capability and performance.

Those categories mix technical evidence with geopolitical judgement. Accuracy can be benchmarked, but censorship and supply-chain risk depend on use context, governance and access. A useful report would need repeatable methods, disclosed limitations and separation between verified vulnerabilities and policy assumptions.

The government already faces related questions around access to advanced chips and cloud compute. Lapaas Voice’s report on AI chip export controls moving toward cloud access shows why model policy cannot be separated from infrastructure controls. Our analysis of the Tencent Hy4 open-model strategy explains how open weights can spread capability even when hardware remains constrained.

What the Open Source AI Bill does not do

The proposal does not create an immediate ban on Chinese models. It does not require American model developers to release proprietary systems. It also does not make Commerce the safety regulator for every open model.

A rule of construction in the introduced text says the act should not be read as authorising the secretary to ban or restrict open models in commerce. That limit matters because the bill’s foreign-risk reports could influence procurement and reputation without becoming a direct prohibition.

The Open Source AI Bill is an adoption-and-transparency proposal, not a model-release mandate: it would help Commerce promote qualifying American open models and publish risk comparisons, while leaving actual bans and broad safety regulation to other legal authorities.

Why the bill matters outside the United States

Developers in India and other markets frequently choose between US and Chinese open-weight models based on cost, language support, licence terms, hardware needs and local deployment. A US government comparison framework could shape enterprise procurement far beyond federal agencies.

That influence could improve disclosure if the methodology is credible. It could also produce a biased market signal if nationality substitutes for technical testing. Indian buyers should continue to evaluate model provenance, licence obligations, data flow, security controls, benchmark relevance and total deployment cost.

What happens next?

The next formal test is the full House Energy and Commerce Committee. Watch for amendments to definitions, reporting deadlines, covered-country rules and the secretary’s powers. Funding and staffing will also determine whether the bill creates useful market data or another lightly used federal report.

The official H.R. 10152 record and introduced text provide the primary legal source. Representative Evans’ subcommittee announcement confirms the September 1 vote and sponsor’s rationale.

Four implementation tests will decide whether it works

The first test is whether Commerce can define adoption without rewarding downloads that never become useful deployments. Download counts are easy to collect but can be inflated by mirrors, experiments and repeated retrievals. Better measures would include active deployments, developer contribution, enterprise retention, supported languages and the cost of running a model at a stated quality level.

The second test is comparison design. A fair comparison must hold tasks, hardware, precision, context length and safety settings constant. Comparing one model’s cheapest configuration with another model’s most capable configuration would produce a politically attractive chart rather than useful procurement evidence.

The third test is disclosure. Public risk reports should describe the evidence behind each finding, distinguish confirmed behaviour from hypothetical misuse and give developers a path to correct factual errors. Sensitive national-security details may require redaction, but unexplained scores would be hard for researchers and companies to trust.

The fourth test is durability. Model capability and pricing can change within weeks, while federal reports often take months. Commerce would need a repeatable update process and machine-readable data if it wants comparisons to remain useful between annual reports.

Startups will also watch whether programme support is neutral among licences and architectures. Some models permit broad commercial use; others restrict high-volume users or particular applications. Calling both “open” can hide material differences in compliance cost and business freedom.

For government buyers, provenance must go beyond the developer’s headquarters. Training infrastructure, code dependencies, fine-tuning data, model hosts and update channels can cross borders. A nationality label alone cannot replace software supply-chain review.

These tests explain why implementation deserves as much attention as the headline bill. A credible programme could make model choice more transparent and reduce adoption friction. A weak programme could turn a fast-moving technical market into a static geopolitical ranking that buyers quickly ignore.

Congressional oversight will be another practical safeguard. Lawmakers can ask whether Commerce updates weak benchmarks, consults independent researchers and measures benefits for smaller developers rather than only established vendors. Those questions would reveal whether the programme is building a competitive open-model ecosystem or simply producing reports. Until the full committee publishes its next action, readers should treat every implementation detail as proposed policy rather than a settled federal requirement.

FAQs

Is the Open-Source AI Leadership Act law?

No. It passed a House subcommittee and still needs additional committee and chamber action.

Does H.R. 10152 ban Chinese AI models?

No. It requires risk assessment and comparisons, while the introduced text says it does not authorise Commerce to ban open models in commerce.

What is a qualified open model?

The bill uses a legal category centred on models developed and made available by US persons, excluding models tied to covered nations or controlled entities.