OpenAI shelves its 2026 IPO because the safety work is not done

Why it matters

OpenAI has enough private capital to postpone Wall Street. The test is whether Altman uses that time to change safety operations, rather than simply delaying disclosure and liquidity.

Sam Altman ruled out a 2026 initial public offering for OpenAI, telling Fortune that unfinished AI safety work made a listing "an ill-advised moment." "I would say not 2026, yeah," Altman said in the interview, according to Axios. "We got a lot of stuff to do."

The statement closes the listing window OpenAI opened on June 8th, when it confidentially submitted a draft S-1 to the Securities and Exchange Commission. OpenAI said at the time that it had not set a timetable and that some work would be easier to complete while private. Three months later, Altman has identified safety as the work that takes precedence.

Altman has spent the past decade moving OpenAI between institutional forms as its capital requirements expanded. He co-founded OpenAI as a nonprofit research lab in 2015 after running Y Combinator, then helped turn it into the commercial organization behind ChatGPT. OpenAI's October 2025 recapitalization placed its operating arm inside OpenAI Group PBC while leaving control with the nonprofit OpenAI Foundation. The Foundation appoints OpenAI Group's directors and houses the Safety and Security Committee that oversees safety practices across the organization.

That structure gives Altman a governance argument for remaining private: OpenAI can continue training and deploying models without adding quarterly earnings pressure to an already strained safety process. It also postpones the financial disclosure, price discovery and shareholder scrutiny that would accompany a public listing.

A filed S-1 becomes an option, not a schedule

OpenAI's confidential S-1 remains important. Filing gave OpenAI the ability to move quickly if management decided the tradeoffs favored a listing. OpenAI explicitly described that flexibility in June, while warning that the process could take time. Altman's September 12th comments convert that open-ended warning into a clear calendar decision.

The postponement will delay a broad liquidity event for employees and private investors, and it leaves OpenAI without a public valuation set through open trading. It also spares OpenAI from having to explain its safety incidents, infrastructure commitments and model economics through recurring securities filings while those systems are changing quickly.

OpenAI can afford that choice. On March 31st, OpenAI announced $122 billion in committed capital at an $852 billion post-money valuation. Amazon, Nvidia and SoftBank anchored the financing, with Microsoft participating alongside a16z, D. E. Shaw Ventures, MGX, TPG and a long list of institutional investors. OpenAI also expanded an undrawn revolving credit facility to about $4.7 billion.

That private capital does more than pay for chips and data centers. It gives Altman time to postpone the compromises of being public without creating an immediate financing crisis. OpenAI still has to satisfy investors that committed capital is producing durable value, but it does not need to ask public shareholders to underwrite the next phase in 2026.

OpenAI's own March figures show why the eventual offering would attract intense scrutiny. OpenAI said ChatGPT had more than 900 million weekly users and 50 million subscribers, while enterprise products generated over 40% of revenue. OpenAI also said it was producing $2 billion in monthly revenue. Those are company-reported metrics, and a public prospectus would force a more detailed accounting of costs, margins, customer concentration and the contractual obligations behind OpenAI's infrastructure buildout.

Safety becomes an operating constraint

Altman's decision landed on the same day Dario Amodei (@DarioAmodei), Anthropic's co-founder and CEO, called for frontier developers to slow capability gains long enough for safety systems to catch up. Amodei warned that AI agents could be six to 12 months from being capable of coordinating internet-scale attacks, and proposed embedded third-party evaluators with employee-like access to frontier labs.

RuntimeWire reported earlier on September 12th that Amodei's three-step plan depends on industry coordination and agreements between governments, while leaving the actual speed limit undefined. Anthropic has committed to the embedded-evaluator component itself.

The warnings followed the resignation of former Anthropic researcher Jacob Coxon, who said OpenAI and Anthropic were prioritizing the model race over safety. Coxon had worked at both organizations and argued that increasingly autonomous systems could threaten human life by the end of the decade. The claims remain his assessment, but they intensified pressure on both CEOs to show where safety can override commercial schedules.

OpenAI had already begun moving in that direction. RuntimeWire reported on September 6th that chief scientist Jakub Pachocki wanted frontier developers to establish shared triggers for slowing development. The proposal came after OpenAI released GPT-6 Astra, a model that OpenAI said produced reasoning that was harder to monitor, and after an agent containment failure sharpened questions about how quickly safeguards can be built around autonomous systems.

An IPO delay is a concrete commercial consequence, though it does not establish that model development itself will slow. OpenAI can remain private while continuing to train larger systems, release products and compete for enterprise contracts. Altman's explanation will be tested against release cadence, safety reporting and the authority given to outside evaluators. A delayed stock sale carries less weight if OpenAI's operational pace remains unchanged.

Anthropic inherits the public-market decision

Anthropic confidentially filed for an IPO before OpenAI and has continued preparing for a possible 2026 listing. Its position became harder to explain on September 12th: Amodei is asking the industry to slow down while Anthropic remains the leading candidate to take a frontier AI developer public this year.

Anthropic also has substantial private financing. In February, Anthropic raised a $30 billion Series G at a $380 billion post-money valuation, led by GIC and Coatue. Anthropic reported $14 billion in run-rate revenue and more than $2.5 billion in run-rate revenue from Claude Code. Those figures are self-reported, but they give Anthropic room to weigh an IPO as a governance and liquidity decision rather than a simple cash requirement.

Altman has chosen continued private control during a period when OpenAI says its models, agents and infrastructure are becoming more consequential. The decision protects management from public-market deadlines and postpones a historic liquidity event. It also creates a measurable standard for OpenAI's next phase: the extra time has to produce safety controls that can keep pace with the systems Altman is still racing to build.