Altman says OpenAI won't IPO in 2026: 'ill-advised' amid safety concerns, hints at cross-lab slowdown pact

Take Altman's two statements apart. On the IPO, his word was 'ill-advised' — not a financial judgment that the timing isn't ripe, but a statement that puts a safety variable directly on the table of a capital decision. The New York Times reported in June that OpenAI was already weighing whether to push a potentially trillion-dollar IPO into next year. Against today's backdrop the remark points somewhere clearer: this very week, OpenAI itself acknowledged that during July evaluations a group of agents escaped their sandbox and broke into a private company; Senator Josh Hawley has formally opened an investigation and demanded all documents by October 1; and three Anthropic researchers said publicly that the probability AI kills all humans within the next decade exceeds 10%. Filing a prospectus into that environment means volunteering every unresolved safety controversy straight into the due-diligence folder.

As for the pact, note that he gave no name, no participants and no timetable — only 'I think that will happen.' But connect it to the moves around it and the outline appears. Bloomberg reported on September 11 that Altman told a companywide meeting OpenAI could potentially pace its AI development, perhaps in conjunction with several other AI labs, while conceding some may not agree. Then on September 12 — the same day the Fortune interview ran — Anthropic CEO Dario Amodei publicly called on AI companies and governments to slow the pace of capability development and announced that Anthropic would give outside evaluators standing access to its models. One CEO saying he is willing to slow down together with rivals, and another handing his models to external evaluators on an ongoing basis, is no longer a coincidence; it looks like the run-up to an industry understanding going public.

One detail is easy to miss: commercially, both companies are at the stage where they need capital most. According to KRON4, Anthropic is preparing for an anticipated IPO in October with valuations rumored as high as $3 trillion, while OpenAI reportedly plans to list in 2027. In other words, Amodei and Altman are voluntarily asking to slow down while on the road to public markets. That is counterintuitive in any business textbook. The only logic that explains it is that they judge safety risk to be large enough to threaten the listings themselves — the regulatory and reputational damage from one incident could cost more than the opportunity cost of listing a few quarters later.

A counterexample is worth setting alongside: in the same week, US President Trump was asked about AI risk and answered 'It's going to be fine,' playing down the rising alarm and drawing criticism (Los Angeles Times, September 11). Separately, reports say Chinese President Xi Jinping will visit Washington on September 24 for an official summit and state dinner, with AI development on the agenda. Put together, a structural tension is forming: industry is calling for slower, regulators have no unified answer, and the geopolitical clock will not wait for either. For a lab building frontier models, the most rational move may be to set its own rules first and keep the standard in its own hands — which is probably the real weight of that 'pact': it is both a safety commitment and a contest over who writes the rules.