OpenAI CEO Altman backs AI pacing, says safety cases will precede frontier reinforcement learning
On September 14 Altman posted on X responding to Amodei's essay, committing OpenAI to drafting explicit safety cases before frontier reinforcement-learning runs expected to substantially increase capability and granting independent evaluators employee-level access.
In the early hours of September 14, 2026, OpenAI CEO Sam Altman posted on X responding to Anthropic CEO Dario Amodei's September 12 essay urging the AI industry to slow the pace of frontier capability gains. Altman wrote that OpenAI welcomes a federal framework setting uniform safety requirements for frontier AI but that companies do not need to wait for an antitrust exemption or new legislation to begin building that trust. Altman committed OpenAI to drafting explicit safety cases in advance of every frontier reinforcement-learning run the company expects to substantially increase AI capabilities, extending safety work beyond the checks OpenAI has historically done before model releases. OpenAI also committed to giving independent evaluators access comparable to that of employees, matching Anthropic's pledge of permanent employee-level access so external reviewers can verify safety practices, report incidents, and assess alignment during training. Altman flagged two long-term risks he said OpenAI weighs in major decisions: loss of human control over AI and concentration of capability in a single company, person, or country. He said OpenAI, Anthropic, xAI, and SpaceX AI have reached broad consensus on embedded evaluators and cross-company safety coordination, and that peer review of AI by competitors is the right starting point for both pacing and shared governance. Hours before Altman's post, Anthropic had already announced it would unilaterally implement the first step of its three-part plan by opening long-term employee-level access inside Anthropic to embedded third-party evaluators.