OpenAI Endorses Four California AI Bills, Calls for Mandatory US Safety Law

OpenAI has formally endorsed four California AI safety bills and is calling on the United States Congress to pass mandatory, capability-based national AI regulation before it adjourns. The company published the positions in a policy post titled "The AI policy window is open. We need to act," arguing that a jump in capabilities has made voluntary commitments insufficient.

The four California bills — SB 813, AB 1405, SB 1119, and AB 1864 — have passed the state legislature and now head to Governor Gavin Newsom's desk. OpenAI says they address complementary parts of the AI-safety ecosystem and can serve as a de facto national baseline until Congress acts.

What is OpenAI asking Congress to do?

OpenAI says the United States needs "mandatory, capability-based national regulation that can evolve as the technology does." The company's Blueprint for Democratic Governance of Frontier AI, which it references, proposes common testing and independent-assessment requirements, stronger cybersecurity protections, clear incident-reporting rules, greater national preparedness, and shared measures for tracking progress toward recursive self-improvement.

OpenAI frames the urgency in stark terms. "With stakes this high, we cannot let the perfect become the enemy of the good," the company wrote. "Congress should act before it adjourns."

The company argues a federal framework should be carefully targeted. Frontier safety requirements, OpenAI says, should apply to "the handful of well-resourced laboratories developing the most capable systems — not to startups, small developers, or researchers operating nowhere near the frontier." Obligations should be proportionate to capabilities and risks.

OpenAI also cautions against turning frontier safety policy into open-weights policy by another name. It notes that it signed the Open Weights and American AI Leadership letter affirming that "America needs both open and closed models." A federal framework, the company argues, should address frontier capabilities without weakening competition, entrenching incumbents, or driving innovation overseas.

Why does the timing matter?

The post ties the policy push to a recent shift in capabilities. OpenAI cites Astra's capabilities, what it calls "early evidence of AI-driven research acceleration," and an essay by Chief Scientist Jakub Pachocki, who wrote that the rapid rise of machine intelligence, including the potential of recursive self-improvement, calls for "extreme caution."

Pachocki's argument, and the policy post built around it, marks a notable position: OpenAI explicitly says it will slow or stop development when necessary. "When proceeding would pose an unacceptable safety risk, we will slow or stop the development or deployment of systems we cannot sufficiently safeguard, as we have done before and as required per our preparedness framework," the company wrote.

The company also invokes what OpenAI President Greg Brockman has described as a "defenders window" — a limited period when frontier AI can help defenders strengthen critical systems before powerful offensive capabilities become widespread. Policymakers face an analogous closing window, the post argues, to establish durable safeguards before AI capabilities outpace the institutions responsible for governing them.

OpenAI's own research, cited in the post, shows that AI agents can now perform some tasks that would take skilled researchers several days. The company is careful to distinguish this from recursive self-improvement — fully autonomous AI-driven AI development, which it says "is not happening today" and should not be pursued "unless and until it can be done safely." But it calls the acceleration "evidence of the direction of travel."

What do the four California bills do?

The bills OpenAI endorsed on Friday cover independent assessments, auditor standards, youth protections, and biosecurity:

• SB 813 would establish a process for designating qualified, independent organizations capable of assessing AI risks. OpenAI says it prefers this function at the federal level, where policymakers can set consistent standards for assessor qualifications, security, and access to highly sensitive information.
• AB 1405 would create registration, independence, transparency, and accountability requirements for AI auditors.
• SB 1119 would require age assurance, risk assessments, independent audits, parental controls, and safeguards against harmful content for children and teens using companion chatbots. OpenAI previously signaled support for the bill, which aligns with protections built into ChatGPT for Teens, including stronger defaults in sensitive areas, Quiet Hours, Study Hours, and break reminders.
• AB 1864 would require gene-synthesis providers and manufacturers of benchtop synthesis equipment to follow federal screening standards, strengthening a physical safeguard against AI-enabled biological threats.

OpenAI disclosed that it did not endorse some of these bills in the past and is supporting them now "after reconsidering in light of the recent jump in capabilities we have seen." The company has previously backed California's SB 53, New York's RAISE Act, Illinois's SB 315, and independent audits in Massachusetts frontier safety legislation — an approach it calls "reverse federalism," in which converging state standards create a baseline Congress can ultimately codify.

The company says the bills are not a substitute for federal regulation but can "protect people now, demonstrate what workable safeguards look like, and help build momentum for federal action."

What does OpenAI want from the industry itself?

Beyond legislation, OpenAI says it will work with other frontier labs to advance voluntary industry standards for frontier AI — "with or without government support." The most urgent starting point, the company argues, is monitoring, particularly for misalignment, which it defines as a model pursuing an objective in ways that violate human intent or established boundaries, without requiring consciousness or malicious intent.

OpenAI says companies should be required to provide prompt written notice to affected parties when their models circumvent another organization's security controls without authorization and materially access, alter, or destroy protected systems or confidential information. It also supports federal reporting requirements for other serious AI incidents.

The company disclosed that it is developing a framework for reporting consequential misalignment incidents and systematically monitoring frontier-model activity, including internal use. It says the effort begins as a company-led initiative but could inform broader federal policy.

Internationally, OpenAI says it will advocate for compatible approaches to measuring capabilities, managing risk, preserving human control, and determining when development should slow or stop — "even if that means slowing the advancement of model capabilities." It argues that credible US standards at home are a precondition for leading internationally.

What has OpenAI changed internally?

The post lists concrete safeguards OpenAI says it has strengthened across the model-development lifecycle:

• Stronger isolation for frontier research workloads
• Expanded monitoring of model behavior during tool-enabled training and evaluations
• Clearer rules for when to escalate concerns
• For Astra, universal monitoring of full trajectories, including chains of thought, and a mandatory alignment-evaluation gate before broader internal deployment

The company's stated aim is to build automated AI researchers that work under human supervision, using each generation of AI to make the next one "safer, more aligned, and easier to control, not simply more capable." Governments, OpenAI argues, should develop common ways to measure that progress and establish shared safety bars. "If we cannot meet certain safety bars without slowing down capability growth, we should prioritize the former," the post states.

The stakes, in OpenAI's telling, run in both directions. Advanced AI could accelerate new medicines, strengthen critical infrastructure, and expand economic opportunity — but the capabilities that make models more useful carry risks that "will not remain confined to a few frontier laboratories," as open models developed around the world increasingly approach today's frontier.

The post closes with a warning about inaction: "No first step will be perfect. But the greater risk now is waiting too long to take one." With Governor Newsom's signature pending on four bills, a pre-adjournment Congress weighing frontier safety proposals, and OpenAI volunteering to slow its own development under conditions it has defined, the next few months will test whether the policy window the company describes gets used — or closes.