WHAT WAS SAID
● Altman, asked directly whether 2026 was still in play: "I would say not 2026."
● On why: an IPO now would be an ill-advised moment given safety concerns, and the company feels no pressure.
● On 2027: he declined to confirm. The CFO has reportedly told associates 2027.
● Same day: Amodei called for slower capability gains. Altman and Musk both agreed publicly.
The window we have been tracking is closed
OpenAI filed confidentially on 8 June, targeting September. We have carried that in our September tracker as targeted, not dated — because the public prospectus never appeared on EDGAR, and registration must be public at least fifteen days before a roadshow.
Each day without a filing narrowed the window. It is now shut, and the reason given is not the one most people assumed.
THE STATED REASON IS SAFETY, NOT MARKETS
Altman was asked directly whether IPO plans were pushing OpenAI to move faster than it should. He pushed back, and framed the delay as sequencing rather than retreat: society needs to contend with these models at each level of capability.
Whether you take that at face value is a judgement. What is not in doubt is that it is the reason given, on the record, to Fortune's editor-in-chief.
The more unusual event happened the same day
Dario Amodei published an essay urging AI companies to slow how quickly they improve their most advanced models, proposing that frontier capability gains be paced to let safety work catch up — and starting with permanent access for third-party evaluators at Anthropic.
Altman and Musk both backed it publicly. Three companies that compete directly, and in some cases litigate against each other, agreeing on a proposal to slow down.
Altman separately told Fortune that OpenAI has discussed pausing development as models cross capability thresholds, and that a formal joint commitment between labs may be close — though he declined to pre-announce private discussions.
Amodei's proposal is not abstract
The permanent third-party evaluator access he proposed is something Anthropic has already started doing. Last week it gave METR wide access to scan millions of evaluation and production transcripts, after disclosing four real-system breaches during evaluations.
That matters because it separates this from the usual safety statement. A proposal costs nothing. A proposal you have already implemented costs something, and it is the version other labs can be asked to match.
What the METR arrangement actually involves →
The tension nobody resolved
| The statement |
The thing that does not pause |
| Slow capability gains to let safety catch up | Compute build-out and spending commitments |
| No pressure to list | A cost base sized for a company that lists |
| Pausing at capability thresholds | GPT-6 Astra shipped ten days ago |
None of those are contradictions exactly. A company can slow releases while continuing to build capacity, and Astra shipped with its cyber capability gated rather than removed. But the spending does not pause for a safety review, and that is the part worth watching rather than the essays.
What actually changes
- Nothing operational. No model, price or availability is affected by any of this.
- For our tracker, the September window moves from targeted to closed, and 2027 becomes the reported expectation rather than a confirmed date. Altman declined to confirm it.
- For prediction markets, this is a resolution event on 2026 listing questions and it pushes everything to 2027 or later.
- For anyone building on OpenAI, a private company answers to different people than a public one. Whether that is better for you depends on whether you would rather your vendor optimised for quarterly results or did not.
Sources
FAQ
Is the OpenAI IPO happening in 2026?
No. Altman told Fortune on 12 September, "I would say not 2026," calling a listing now an ill-advised moment given safety concerns.
Is 2027 confirmed?
No. Altman declined to confirm a specific year. CFO Sarah Friar has reportedly told associates the company is targeting 2027, and she previously told employees 2027 or sooner if the business continued to inflect.
What reason did he give?
Safety and alignment work, and how industry and governments can coordinate. He pushed back directly on the suggestion that IPO plans were making OpenAI move faster than it should.
What did Amodei propose?
Pacing frontier capability gains so safety work can catch up, starting with permanent access for third-party evaluators at Anthropic. Altman and Musk both backed it publicly.
Does this affect me as a user or developer?
Not operationally. No model, pricing or availability changes as a result.
Had OpenAI already filed?
It filed confidentially on 8 June. The public prospectus never appeared on EDGAR, and registration must be public at least fifteen days before a roadshow, which is why the September window was never a confirmed date.