Chris Lehane, Chief Global Affairs Officer at OpenAI, argues that rapid advances in AI capabilities require new policy approaches. Lehane says no single company, industry, or government can handle the risks alone and that policymakers should prioritize concrete action over waiting for perfect solutions.
Why OpenAI sees this as urgent
OpenAI notes that advanced AI could deliver significant benefits—accelerating drug discovery, strengthening critical infrastructure, expanding economic opportunity, and helping solve long-standing scientific problems—but also carries substantial risks. The company warns that these capabilities will not remain limited to a few frontier labs; models developed worldwide, including open models, will increasingly approach today’s frontier and become broadly available.
Lehane cites Jakub Pachocki’s call for “extreme caution” and references Greg Brockman’s “defenders window” idea: a limited period during which defenders can harden systems before offensive capabilities become widespread. Policymakers face a similar closing window to establish durable safeguards.
Technical and organizational measures OpenAI describes
OpenAI says it has strengthened monitoring, alignment, and security across the model-development lifecycle: stronger isolation for frontier research workloads, expanded monitoring of model behavior during tool-enabled training and evaluations, and clearer rules for escalating concerns. For the system called Astra, the company introduced universal monitoring of full trajectories (including chains of thought) and a mandatory alignment-evaluation gate before broader internal deployment.
The company states that it will slow or stop development or deployment when proceeding would pose an unacceptable safety risk, as it has done before and as required by its preparedness framework.
OpenAI emphasizes that fully autonomous recursive self-improvement—AI systems independently producing successive, more capable generations—does not exist today and should not be pursued unless it can be done safely. At the same time, AI is already accelerating parts of the research used to build and align next-generation models: OpenAI’s recent research shows AI agents can perform some tasks that would take skilled researchers several days.
Their stated aim is to develop automated AI researchers that operate under human supervision to advance both deep learning and alignment—using each generation of AI to help make the next one safer, more aligned, and easier to control rather than merely more capable.
Proposal for mandatory, capability-based federal rules
OpenAI calls for mandatory, capability-based national regulation in the United States that can evolve with the technology. The company’s Blueprint for Democratic Governance of Frontier AI outlines a federal framework including common testing and independent-assessment requirements, stronger cybersecurity protections, clear incident-reporting rules, greater national preparedness, and shared measures to track progress toward recursive self-improvement.
OpenAI says frontier safety requirements should target the relatively small number of well-resourced laboratories developing the most capable systems, not startups, small developers, or researchers who are far from the frontier. Obligations should be proportionate to capabilities and risks.
The company also warns against letting frontier safety policy become a de facto “open-weights” policy. Open models can contribute to solutions—particularly in cybersecurity or where sovereignty, security, or data-residency needs favor local deployment—but most open models compete on cost, control, and latency rather than at the frontier. OpenAI notes that it signed the Open Weights and American AI Leadership letter and believes the U.S. needs both open and closed models. A federal framework should address frontier capabilities and risks without weakening competition, entrenching incumbents, or driving innovation overseas.
OpenAI argues a public framework should reduce concentration of power by replacing the current fragmented private governance—where frontier labs largely set their own rules—with democratically accountable standards, independent verification, and meaningful transparency.
State-level steps and specific bills OpenAI supports
Until Congress acts, OpenAI says states should continue to raise the safety baseline. The company previously supported California SB 53, New York’s RAISE Act, Illinois SB 315, and independent audits included in frontier safety legislation under consideration in Massachusetts. It encourages states to converge on common safeguards to create a de facto national baseline that Congress can later codify (an approach the company calls reverse federalism).
OpenAI now formally endorses four California bills that have passed the legislature and are headed to Governor Gavin Newsom:
-
SB 813: establishes a process to designate qualified, independent organizations capable of assessing AI risks. OpenAI prefers federally required independent technical assessments but says California can help set national rules in the absence of federal action.
-
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 companion chatbots used by children and teens. OpenAI notes prior support for youth-safety measures, including product controls in ChatGPT for Teens.
-
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 risks.
Lehane notes that OpenAI did not support some of these bills in the past but has reconsidered in light of recent capability advances. He reiterates that state bills are not a substitute for federal regulation but can provide immediate protection and demonstrate workable safeguards.
Monitoring, incident reporting, and industry standards
OpenAI stresses that laws alone will not make advanced AI safe and that concrete laboratory practices—particularly monitoring—are urgently needed. Monitoring is crucial to detect misalignment, when a model pursues objectives in ways that violate human intent or established boundaries even without consciousness or malice.
Developers should be required to monitor for misaligned behaviors and demonstrate effective safeguards. Monitoring must be linked to clear disclosure requirements: OpenAI supports prompt written notice to affected parties when, during development or evaluation, a model circumvents another organization’s security controls without authorization and materially accesses, alters, or destroys protected systems or confidential information. The company also supports federal reporting requirements for other serious AI incidents and is working to define which incidents should be covered and what the requirements should entail.
OpenAI says it is developing a framework for reporting consequential misalignment incidents and systematically monitoring frontier-model activity, including internal use. That effort begins as a company-led initiative but could inform broader federal policy.
The company wants to work with other labs on industry-led standards that would complement—but not replace—mandatory federal safeguards and democratic oversight. OpenAI also stresses that any standards must ultimately extend beyond national borders, because models, research, and technical expertise move globally and failures with the most capable systems could have international consequences.
Conclusion
OpenAI concludes that the policy window to act is open for now and must be used with urgency, humility, and readiness to adapt. The company says it will support serious proposals that materially raise the safety bar even if they are not exactly what OpenAI would design, and it calls on Congress to act before it adjourns. The greater risk, Lehane argues, is waiting too long to take the first meaningful steps.
