- Potential benefitGenerates empirical, test‑based data to inform Congress and regulators, enabling more targeted and evidence‑based AI ov…
- Potential benefitImproves national security and public safety by systematically identifying weaponization risks, loss‑of‑control scenari…
- Federal agenciesCreates government and contractor jobs (testing engineers, red teams, analysts, security and handling personnel) and co…
Artificial Intelligence Risk Evaluation Act of 2025
Read twice and referred to the Committee on Commerce, Science, and Transportation.
The bill establishes an Advanced Artificial Intelligence Evaluation Program at the Department of Energy. It requires covered developers of specified “advanced artificial intelligence systems” to participate in testing, provide materials (potentially including code, model weights, and training data) on request, and prohibits deployment in interstate or foreign commerce unless in compliance, with penalties for violations.
Compelled disclosure vs.
Relative to its intended legislative type, this bill establishes substantive legal obligations and prohibitions and creates a technically ambitious federal evaluation program with statutory timelines and reporting requirements.
The bill establishes an Advanced Artificial Intelligence Evaluation Program at the Department of Energy.
It requires covered developers of specified “advanced artificial intelligence systems” to participate in testing, provide materials (potentially including code, model weights, and training data) on request, and prohibits deployment in interstate or foreign commerce unless in compliance, with penalties for violations.
The program would run classified and standardized red-team testing, produce reports and risk assessments, recommend containment and mitigation strategies, and deliver a proposed permanent federal oversight framework within 360 days, including options up to nationalization if ‘‘artificial superintelligence’’ appears likely.
On content alone, the bill is ambitious and intrusive: it imposes mandatory disclosures, criminalizes deployment absent compliance, contemplates drastic remedies (including nationalization), and creates a substantial new federal testing apparatus without explicit funding language. These features make it politically and practically difficult to advance without major modification. Elements that could help—national security framing, bipartisan concern about AI risks, and use of existing DOE lab capabilities—are unlikely to overcome opposition to compelled disclosure and heavy penalties without negotiated compromises.
Relative to its intended legislative type, this bill establishes substantive legal obligations and prohibitions and creates a technically ambitious federal evaluation program with statutory timelines and reporting requirements. It clearly sets out purposes and several program functions but leaves substantial implementation, fiscal, procedural, and legal details unspecified.
Compelled disclosure vs. IP/privacy protections: liberals/centrists see disclosure as necessary for risk assessment (with safeguards); conservatives see it as a threat to trade secrets and competitiveness.
Who stands to gain, and who may push back.
These are examples from the analysis, not a ranked list of the most-affected groups.
- DevelopersImposes substantial compliance costs and operational burdens on AI developers (sharing of code, weights, training data,…
- Federal agenciesRaises trade‑secret and privacy concerns because mandatory disclosure of underlying models and training data to a feder…
- Potential burdenCreates significant regulatory uncertainty and potential chilling effects from the deployment prohibition and steep civ…
Why the argument around this bill splits.
Compelled disclosure vs. IP/privacy protections: liberals/centrists see disclosure as necessary for risk assessment (with safeguards); conservatives see it as a threat to trade secrets and competitiveness.
A mainstream liberal/left-leaning observer would likely view the bill positively as a robust, government-led effort to gather empirical evidence and manage the social, civil-liberties, labor-market, and existential risks posed by advanced AI.
They would welcome the focus on red-teaming, classified testing (where needed), and requirements to study impacts on civil liberties and labor markets.
However, they would also want safeguards for privacy, protections for marginalized groups, and transparency about how private data and proprietary information are handled.
A centrist/moderate observer would probably be cautiously supportive of the bill’s core idea—creating an empirical, technical testing program to inform regulation—but would be concerned about ambiguities and implementation details.
They would praise the DOE placement (given DOE’s supercomputing and lab expertise) and the law’s aim to base oversight on data, while worrying about scope, definition clarity, costs, legal authority to compel proprietary material, and the $1,000,000-per-day fine.
The centrist would seek clearer thresholds, procedural protections for IP and privacy, and phased or risk-based obligations to reduce burdens on smaller actors.
A mainstream conservative observer would likely view the bill skeptically as a significant expansion of federal power over private-sector technology, with compulsory disclosure of proprietary materials, a deployment ban tied to compliance, and the explicit option of nationalization.
They would be concerned about the impact on innovation, competitiveness, intellectual property rights, and commercial confidentiality, and about placing a politically appointed official (Secretary of Energy) in a gatekeeping role for market access.
While recognizing national-security rationales for some testing, they would prefer industry-driven standards, limited government access, and clearer protections for U.S. competitiveness.
The path through Congress.
Reached or meaningfully advanced
Reached or meaningfully advanced
Still ahead
Still ahead
Still ahead
On content alone, the bill is ambitious and intrusive: it imposes mandatory disclosures, criminalizes deployment absent compliance, contemplates drastic remedies (including nationalization), and creates a substantial new federal testing apparatus without explicit funding language. These features make it politically and practically difficult to advance without major modification. Elements that could help—national security framing, bipartisan concern about AI risks, and use of existing DOE lab capabilities—are unlikely to overcome opposition to compelled disclosure and heavy penalties without negotiated compromises.
- The bill's definition of "advanced" AI relies on a computing‑power threshold that is unclear in the text (formatting suggests a large exponent but is not explicit); this ambiguity affects who is covered and thus the practical scope.
- No explicit appropriations or funding mechanism is included; cost and resource needs for classified testing and ongoing evaluations are unknown and would affect political feasibility.
Recent votes on the bill.
No vote history yet
The bill has not accumulated any surfaced votes yet.
Go deeper than the headline read.
Compelled disclosure vs. IP/privacy protections: liberals/centrists see disclosure as necessary for risk assessment (with safeguards); cons…
On content alone, the bill is ambitious and intrusive: it imposes mandatory disclosures, criminalizes deployment absent compliance, contemp…
Relative to its intended legislative type, this bill establishes substantive legal obligations and prohibitions and creates a technically ambitious federal evaluation program with statutory timelines and reporting requi…
Go beyond the headline summary with full stakeholder mapping, legislative design analysis, passage barriers, and lens-by-lens tradeoff breakdowns.