OpenAI delays its IPO as Altman, Amodei and Musk call for slowing down AI in a rare united front

By Zendoric · September 13, 2026.

Sam Altman has ruled out an OpenAI stock market listing this year. In an interview published Saturday by Fortune, the chief executive called a public offering now "ill-advised," pushing back to 2027 at the earliest one of the most anticipated IPOs in tech history. Barely a month ago, OpenAI's chief financial officer, Sarah Friar, had told staff the company would probably go public in 2027 "or even sooner" if the business kept "inflecting": Altman's turnaround comes fast.

That same Saturday, Dario Amodei, chief executive of Anthropic, published an essay calling on AI companies to deliberately set the pace at which they improve their most powerful models. He proposes a three-step plan: that Anthropic —which has already done so, unilaterally— give independent external evaluators "employee-level" access to verify its safety practices and report incidents; that the major AI companies in democratic countries coordinate common safety standards; and that, further down the line, there be coordination between democratic and authoritarian governments. Amodei qualifies: "pacing" does not mean halting training or technical progress, but allowing enough time to align and secure models before releasing them, with external verification.

The reaction was immediate. On X, Sam Altman said he agreed with Amodei and confirmed OpenAI will do the same with independent evaluators granted "employee-like" access ("we'll have more to share soon"), explaining that the topic has been "a major subject" of internal discussion in recent weeks. Elon Musk, whose xAI was absorbed this year by his own SpaceX, wrote simply: "Dario is right." It is a notable shift: Musk had previously accused Anthropic of "hating Western civilization" and of being "doomed to become the opposite of its name" (misanthropic); the change in tone coincides with a computing deal between SpaceX and Anthropic closed in May, after which Musk said he had not detected "evil" in anyone at the company.

Amodei's essay does not appear in a vacuum. Earlier this month, OpenAI's chief scientist, Jakub Pachocki, had already warned in a post that no AI company has "solved alignment and monitoring well enough" to keep scaling responsibly at full speed for much longer, and said he expects voluntary pauses to become "commonplace" until shared safety standards exist. And this week, a researcher who has worked at both Anthropic and OpenAI, Jacob Coxon, resigned, claiming that both companies are "gambling with our lives"; according to Coxon, those building these systems genuinely believe AI could kill us all before the decade is out. These are extreme claims, attributable only to the person making them, but they are not new: in 2023 Amodei and Altman already signed, alongside other leading researchers, a statement placing the risk of extinction from AI on a par with pandemics or nuclear war.

In Washington, lawmakers from both parties are calling for new safeguards and hearings after a series of cyberattacks carried out without direct human control. At the state and local level, opposition is growing to AI data centers over their electricity and water consumption and their impact on communities. With the midterm elections around the corner, AI is becoming a campaign issue for both parties.

Our read: that three such bitter rivals —Altman, Amodei and Musk have been trading public jabs for years— should converge on the same day in calling for a brake is the strongest sign yet that the safety conversation has stopped being a niche debate among researchers and has entered the room where the product calendar and the financial calendar are decided. The data point that really carries weight here is not Amodei's essay —essays are cheap— but that OpenAI is shelving a historic IPO. Delaying a deal of that magnitude carries a real opportunity cost; committing to external audits in a blog post, not so much. It is worth measuring commitment by what it costs, not by what is announced.

That said, it is worth separating the verifiable from the aspirational within Amodei's own plan. The first step —giving external evaluators "employee-level" access— is concrete, is already under way at Anthropic and is now promised by OpenAI too: it is auditable, and compliance can be checked. The second and third steps —common standards among democratic companies, and then coordination with authoritarian governments— are, for now, pure aspiration, with no mechanism or timetable, and Amodei himself admits that "some steps will be easier than others." It is the same pattern we see again and again in this sector: the demonstrated and the promised are announced in the same paragraph, and only the former deserves immediate credit.

There is also a tension the announcement itself does not resolve: the company calling loudest for a slower pace is, at the same time, the one actively preparing for what CNBC itself describes as a "historic" IPO. Amodei clarifies that "pacing" is not "stopping," and it is a real distinction —nothing in the essay implies Anthropic will delay its own releases— but it is worth watching whether that pace translates into anything more than extra audits before each commercial launch. Safety and a stock market timetable need not clash, but neither should we assume they won't.

At Zendoric we have spent months arguing that the AI problem is not black or white: in the short term there is real friction —we saw it raw this week, with a public resignation invoking extinction risk and with cyberattacks already carried out without direct human intervention— and in the long term the technology's potential to cure diseases, extend healthy life and generate abundance remains intact. Nothing published this Saturday contradicts that thesis; on the contrary, it reinforces it: that the sector's own leaders publicly acknowledge that time is needed to "align and secure" models is exactly the kind of discipline that separates a well-managed future of abundance from an avoidable accident. The question that really matters from here on is not whether Altman, Amodei and Musk said the right words on the same Saturday —that has already happened— but whether, six months from now, there are external evaluators with real access, published audits and some launch genuinely delayed for safety reasons. That will be the data point that confirms, or refutes, everything announced this week.