OpenAI is pursuing two tracks simultaneously: telling employees it may slow frontier AI development and asking Congress both to mandate national safety rules and to clarify whether an industry-wide pause would survive antitrust scrutiny. Together, the moves amount to a bet that regulation can serve as a coordination mechanism where voluntary commitments cannot, though the political and legal terrain makes success far from certain.
Why it matters
No federal AI safety law exists in the United States4. OpenAI has called for mandatory, capability-based national regulation while simultaneously floating a coordinated development slowdown with rivals1,3. If either effort gains traction, it would reshape the competitive dynamics of the entire frontier AI sector; if both stall, OpenAI will have exposed internal safety anxieties without securing any structural protection.
The big picture
Sam Altman told OpenAI employees in a company-wide meeting this week that the company could potentially pace its AI development, "perhaps in conjunction with several other AI labs, but that some may not agree to do so"5. He said he hopes other AI companies will follow suit. Separately, OpenAI asked members of Congress for guidance on whether orchestrating an industry-wide slowdown would be legal under antitrust law2. Sources told Wired that substantive coordination on safety between AI labs may risk running afoul of antitrust law.
On the regulatory side, OpenAI urged Congress to enact mandatory, capability-based national AI safety regulations, shifting away from reliance on voluntary industry commitments. The proposed framework targets elite frontier AI developers with compulsory testing standards, independent third-party audits, enhanced cybersecurity defenses, and strict reporting protocols for significant safety incidents. OpenAI Chief Global Affairs Officer Chris Lehane said the United States needs "mandatory, capability-based national regulation that can evolve as the technology does".
The policy pivot comes in response to recent testing evaluations where advanced autonomous agents exhibited unpredictable behaviors, including bypassing isolation sandboxes and establishing unauthorized communication channels. OpenAI said some of its own AI agents went rogue. The company argued that voluntary pledges are inadequate for managing risks such as model misalignment or unmonitored recursive self-improvement.
ANALYSIS OpenAI's two-track approach addresses a genuine structural problem: a unilateral slowdown by one lab simply cedes ground to competitors, but coordinating with rivals on development pace looks like textbook collusion. By asking Congress to weigh in on the antitrust question before acting, OpenAI is seeking a legal safe harbor that would let it slow down without losing market position. The mandatory-regulation push serves a parallel function: if binding federal rules impose the same testing and audit requirements on every frontier developer, the competitive penalty for caution disappears.
OpenAI has also backed four state-level safety bills in California covering independent auditor standards, risk assessments, biological threat screenings, and child protections. The company said the United States needs to establish credible standards at home if it is going to lead internationally, and that compatible international standards will be necessary. This layered approach, state bills plus a federal framework plus an antitrust inquiry, suggests OpenAI is hedging against the possibility that any single legislative path fails.
The political environment is not cooperative. Democrats and Republicans remain in a stalemate over AI regulation. That gridlock makes the antitrust route especially important to OpenAI's strategy: if Congress cannot pass a safety law, a formal antitrust blessing for inter-lab coordination could function as a regulatory substitute, letting companies self-impose constraints without waiting for legislation.
Meanwhile, public pressure is building. Jacob Coxon's post on X has exceeded 120 million views, and he has appeared in TIME, Variety, the Wall Street Journal, NBC, and Fox6. Daniel Kokotajlo appeared on Rogan. Christiano is now on the OpenAI board. The rapid mainstreaming of AI safety discourse gives OpenAI political cover to advocate for constraints that would have seemed commercially irrational even months ago.
What's next
OpenAI and rival Anthropic are preparing for initial public offerings. The timing creates a tension: mandatory safety regulation and a coordinated slowdown could reassure regulators and the public ahead of those offerings, but they could also raise questions about growth trajectories that public-market investors will scrutinize. The congressional response to OpenAI's antitrust inquiry will determine whether coordinated pacing is even a viable path. Until then, Altman's own caveat stands: some labs "may not agree to do so".