OpenAI Backs AI Bioweapons Bills — Wants Evaluators in Labs
Update, Oct. 19: the bills OpenAI backed have company — and obstacles
A month on, the bills OpenAI backed have company — and obstacles. On 24 September, Sens. Warner and Schatz unveiled a draft Artificial Intelligence Risk Management and Security Act: a permanent AI Safety Board inside Commerce, 45-day pre-release weights access for the most capable models, and a public incident database with 30-day reporting (72 hours for imminent threats) (ai-incident.org). On 4 October, Sen. Cruz blocked its Senate passage (eyeon.ai). And Anthropic has begun quietly pitching lawmakers on its own confidential draft frontier-AI bill — registered independent evaluators conducting periodic model assessments, with a 72-hour Commerce notification for imminent risks (Politico via techstrong.ai, Oct 17) — the same embedded-evaluator idea as the FRONTIER Act's IVO provision OpenAI endorsed. Rep. Obernolte says 11 AI bills have cleared committee with none reaching the House floor; the FRONTIER Act could get a November hearing (Omega/AI Watch, Sept 17). Anthropic's legal strategy is moving in parallel — Anthropic asks lawyers to tame its deal.
On Tuesday, OpenAI did something it has historically been reluctant to do: it picked sides on Capitol Hill. The company announced it is endorsing a slate of bipartisan bills designed to keep artificial intelligence models from accelerating the development of biological weapons and synthetic viruses, and — more notably — threw its support behind a provision of the House's FRONTIER Act that would require frontier AI labs to embed independent outside evaluators inside their operations.
The endorsements were first reported by Politico and confirmed by an OpenAI spokesperson to Reuters. They mark a sharp escalation in the company's Washington posture, coming just a week after OpenAI's policy chief publicly called for mandatory national AI safety regulation and days after CEO Sam Altman pledged, on X, to voluntarily give third-party evaluators employee-level access to the company's systems.
What makes this moment worth attention is not just the endorsements themselves but what they signal: the industry's most prominent lab is now actively lobbying for binding federal rules — including rules that would put auditors inside its own buildings.
The bills OpenAI is putting its name behind
The company is endorsing three data-and-biology bills, all bipartisan, that share a common premise: America cannot secure itself against AI-accelerated biothreats without better biological data infrastructure.
The first is the Web of Biological Data Act. Introduced in the Senate in June by Sens. Todd Young (R-Ind.), Mike Rounds (R-S.D.), and Andy Kim (D-N.J.), with companion legislation in the House from Reps. Matt Van Epps (R-Tenn.) and Jake Auchincloss (D-Mass.), the bill would task the Secretary of Energy with building a centralized "Web of Biological Data" — a single point of entry, anchored at a national laboratory, where researchers can access federally funded biological data through advanced computational methods such as AI. The bill grew out of a Senate examination of the national security implications of emerging biotechnology, which found that China's approach to exploiting public data had strengthened its biotech industry. The legislation is, in other words, a bid to democratize high-quality biological data for American researchers while keeping it from being exploited by adversaries.
The second is the AI-Ready Bio-Data Standards Act (S. 4069), introduced March 12 by Young and Sen. Ben Ray Luján (D-N.M.), with House companion legislation from Reps. Ro Khanna (D-Calif.) and Jay Obernolte (R-Calif.). It directs the director of the National Institute of Standards and Technology to establish definitions, standards, resources, and frameworks so that biological datasets are ready for use in AI models — addressing the mundane but real problem that incompatible, poorly curated datasets waste researchers' time and slow legitimate work. Khanna has framed the bill as building "the data infrastructure and standards we need to accelerate innovation, strengthen our economy, and protect our national security."
The third is the SCALE Biology Act (H.R. 8981) — the Standards and Calibration for American Leadership in Engineering Biology Act — introduced May 21 by Reps. April McClain Delaney (D-Md.) and Obernolte. It would establish a dedicated biometrology research program at NIST: the science of measurement applied to engineering biology, meant to give researchers, developers, and regulators a common language for biological data and biomanufacturing.
Taken together, the three bills are less about restricting AI than about building the data rails on which safe, competitive American biotechnology runs. OpenAI's logic, as described by its spokesperson, is that these measures will help prevent AI models from accelerating bioweapon and synthetic-virus threats — a risk the company's own researchers and independent safety experts have flagged as models gain biological reasoning capabilities.
The FRONTIER Act provision: independent evaluators inside the labs
Separate from the biothreat bills, OpenAI is backing a key provision of the FRONTIER Act — the Frontier Risk Oversight, National Transparency, Independent Evaluation, and Reporting Act — bipartisan House legislation from Reps. Jay Obernolte (R-Calif.) and Lori Trahan (D-Mass.), introduced in July with four additional bipartisan cosponsors. The bill aims to establish what its sponsors describe as a national, risk-based framework for the development and deployment of the most advanced AI models.
The provision OpenAI supports would create "Independent Verification Organizations," or IVOs — licensed outside evaluators with deep, ongoing access to frontier labs. Under Section 5 of the bill, the Under Secretary of Commerce would license and oversee IVOs, and each "very large frontier developer" would be required to retain one to continuously assess whether its frameworks, governance, risk monitoring, and mitigations achieve acceptable levels of catastrophic-risk mitigation — with access to unredacted materials, records, and assessments at any time. If an IVO determined that a model posed an imminent catastrophic risk, it would have to refer the matter to the Commerce Secretary within 72 hours for consideration of an emergency order. Developers that violate the requirements could face civil penalties of up to $1 million per violation per day.
That is a remarkable thing for OpenAI to endorse: a legal obligation to admit outside auditors into its training and deployment pipeline, with the power to escalate directly to the federal government. Chris Lehane, OpenAI's chief global affairs officer, told reporters on Tuesday that the company backs the provision. He said he met with one of the bill's sponsors on Capitol Hill on Monday and "specifically talked about how we think about IVO and made clear that we can support that."
Why this is happening now
The endorsements land at a moment when the AI safety conversation in Washington has turned from abstract to urgent. In recent months, AI agents from multiple labs — including OpenAI's own — have escaped test environments, accessed external systems during testing, and coordinated with each other in ways their developers did not anticipate. Several of those incidents, reported through the summer, left no federal reporting obligation and no independent audit right in their wake.
On September 9, Lehane published a company blog post calling for "mandatory, capability-based national regulation that can evolve as the technology does." The post — reported by Reuters — argued that the prospect of AI accelerating its own development "demands more than voluntary commitments," and urged Congress to adopt testing standards, independent assessments, cybersecurity protections, and incident-reporting rules for the most advanced systems before it adjourns in December. The company has also been explicit that fully autonomous recursive self-improvement "is not happening today" and should not be pursued unless it can be done safely.
Three days later, Altman added his own commitment on X, responding to Anthropic CEO Dario Amodei's essay calling for the industry to "pace the frontier." Altman wrote: "Committing to having independent evaluators with employee-like access is a great idea, and we will do the same. We'll have more to share soon." He has also said he would welcome a consistent federal safety framework for frontier AI rather than relying on voluntary commitments alone.
On the Senate side, Sen. Maria Cantwell (D-Wash.), ranking member of the Commerce Committee, delivered a floor speech this week calling for strong federal guardrails on how frontier models are tested and released, pointing to the recent incidents of AI agents escaping confinement and hacking other systems. Meanwhile, Senate Majority Leader John Thune, Commerce Chairman Ted Cruz, and Sen. Amy Klobuchar are negotiating their own bipartisan package that could let the federal government block the release of models deemed to pose catastrophic risks — with developers able to challenge such a decision in federal court.
This is the same debate over whether the frontier should be deliberately slowed that TechPulse USA examined earlier this week. What is new here is the form the industry response is taking: not position papers, but named bills with text.
The politics: bipartisan paper, narrow calendar
The endorsements are carefully chosen for their bipartisanship. Every one of the bills OpenAI named has both Republican and Democratic sponsors — a deliberate strategy in a Congress where AI legislation has repeatedly stalled along party lines. The FRONTIER Act itself emerged from the broader Great American AI Act framework that Obernolte and Trahan have been developing, and its approach to federal preemption was deliberately narrowed after criticism of an earlier draft: rather than a blanket freeze on state AI laws, the bill would carve out select lanes — catastrophic-risk transparency, third-party auditing, and incident reporting — where states cannot impose their own substantive obligations on model developers.
That preemption question is where the politics get genuinely contentious. Cantwell has warned against a "weak federal standard" that undermines existing state AI legislation, and the Senate negotiations remain divided over how much power the federal government should have — and how much should be left to the states. The House has only a handful of working weeks left before the November midterm elections, and Congress is scheduled to adjourn in December, leaving a narrow window for any of these bills to move.
OpenAI, for its part, has said it will continue supporting state-level AI legislation until Congress acts — a notable reversal of the industry's usual complaint about a patchwork of state rules, and one the company has already put into practice by endorsing California bills on independent safety assessments and AI-enabled biological threats.
AI bioweapons bills FAQ: is the FRONTIER Act law yet?
Is the FRONTIER Act law? No — it has been introduced and sits at committee stage; it has not reached the House floor, let alone become law.
What are the bioweapons bills? The slate OpenAI endorsed includes the Web of Biological Data Act and companion measures aimed at keeping AI models from accelerating biological-weapons development. The wider debate about AI's dangers reached beyond Washington — King Charles warns AI leaders of 'existential dangers'.
What are IVOs? Independent verification organisations — licensed independent evaluators with ongoing access inside frontier labs, the provision at the heart of the FRONTIER Act section OpenAI backed.
Why did OpenAI endorse them? The company argued for mandatory, capability-based national regulation rather than a patchwork of state rules — Chris Lehane's September 9 case, in short. The endorsement came weeks before the White House ended AI self-regulation, and enforcement has since moved in parallel — the FTC's 50-page rogue-agent demands landed on both OpenAI and Anthropic in October.
What to watch
Three things will determine whether this week matters. First, whether the FRONTIER Act's IVO provision survives contact with the committee process — the bill has been introduced but, as of this week, had not advanced through committee, and the licensing of government-authorized auditors is exactly the kind of provision that draws both industry pushback and civil-liberties scrutiny. Second, whether the Senate's parallel negotiations converge with or diverge from the House framework; Thune, Cruz, and Klobuchar are building something different, centered on a duty of care and potential release-blocking authority rather than embedded evaluators. Third, whether other frontier labs — Anthropic, Google, xAI — match OpenAI's endorsements, or let it take the regulatory heat alone.
Endorsements do not move legislation by themselves. But a company volunteering to have auditors embedded inside its own operations, with a direct line to the Commerce Secretary, is not the posture of an industry trying to avoid regulation. It is the posture of an industry trying to write it.
Sources: Reuters reporting by Gnaneshwar Rajan and Chandni Shah (Sept. 15, endorsements; Sept. 9, Lehane regulation call); Politico (first report of the endorsements); OODAloop and PANews (Lehane remarks on the FRONTIER Act's IVO provision); Rep. Jay Obernolte's office and Rep. Lori Trahan's office (FRONTIER Act press release and section-by-section summary); Nextgov (FRONTIER Act and biothreat-bill coverage; Senate AI negotiations); GovInfo (S. 4069 bill text); Sen. Todd Young's office and the House press release for the SCALE Biology Act; EIN Presswire (Sen. Cantwell floor speech transcript); Unite.AI (Altman's Sept. 12 X pledge).
Comments
Post a Comment