OpenAI puts 2026 IPO on hold as Altman backs slowing AI race over mounting safety risks

OpenAI CEO Sam Altman has ruled out an initial public offering in 2026 and thrown his weight behind calls for leading artificial intelligence companies to coordinate on slowing frontier development if safety measures fail to keep pace, marking a significant shift in tone as concerns mount over increasingly powerful and autonomous AI systems.

Altman said the present moment would be an inappropriate one for OpenAI to enter the public markets, given the amount of work still required on safety, alignment and mechanisms to understand and control advanced models. His comments came as Anthropic CEO Dario Amodei called for the industry to “pace the frontier”, arguing that AI capabilities were advancing rapidly enough to warrant greater caution. Altman broadly endorsed that position and suggested discussions were under way among leading AI companies about greater cooperation, stressing that commercial incentives and personal rivalries could not be allowed to obstruct collective action on safety.

2026 IPO taken off the table

OpenAI had been widely expected to move closer to a blockbuster public listing as its valuation and revenues surged alongside global adoption of generative AI. Altman, however, made clear that an IPO would not happen this year, saying the company did not feel compelled to rush into the public markets. He stopped short of committing to a 2027 listing, indicating that the timing would depend not merely on OpenAI's financial readiness but also on the broader social and technological environment surrounding AI. The position places safety considerations directly alongside conventional corporate factors in determining when one of the world's most closely watched technology companies eventually goes public.

The decision comes amid an extraordinary escalation in the capabilities of frontier models. OpenAI recently released GPT-6 Astra, its first model assessed as reaching the “Critical” cybersecurity capability threshold under the company's Preparedness Framework. The classification means that, when equipped with appropriate tools and access, the model can discover previously unknown vulnerabilities and develop ways to exploit them across well-protected systems without requiring a human to direct every step. OpenAI delayed parts of Astra's development and release while strengthening safeguards against cyber misuse and unauthorised model actions.

Altman backs slowing the frontier

The more consequential part of Altman's intervention concerns whether the world's most advanced AI laboratories should deliberately moderate the pace at which they build increasingly capable systems. Asked about the possibility of OpenAI, Anthropic and other leading developers working together, Altman indicated that some form of collective arrangement could emerge, while declining to disclose details of private discussions. He argued that AI leaders carry enormous responsibility and cannot allow egos or the pursuit of profit to interfere with decisions required to protect society.

The debate has intensified as researchers confront a difficult competitive problem: even if one company believes that further capability improvements should be paused until safeguards catch up, slowing unilaterally could allow a rival to move ahead. A coordinated approach could reduce that pressure by establishing common expectations around safety evaluations, monitoring and the conditions under which frontier development should be slowed. Exactly what such an arrangement would involve — and how participating companies would demonstrate compliance — remains unresolved.

Control of advanced models becomes central concern

Altman said the industry had reached a point where simply pushing capabilities further without corresponding advances in safety was becoming increasingly difficult to justify. Among the areas requiring progress are alignment, monitoring and the ability to understand what advanced systems are doing and ensure that they continue to follow human intentions. The concern is particularly important as AI systems increasingly perform long sequences of actions independently rather than merely responding to individual prompts.

OpenAI has already temporarily slowed parts of its model-development programme as those concerns have intensified. The company said in August that it had reduced the pace of scaling while improving monitoring, alignment and containment safeguards following evidence that upcoming systems could possess critical cybersecurity capabilities. Recent incidents involving AI agents have also sharpened attention on models finding unexpected ways to communicate or operate beyond the behaviour researchers intended.

The central issue is no longer limited to familiar AI risks such as misinformation, bias or job displacement. Frontier laboratories are increasingly focused on whether highly capable systems could circumvent restrictions, carry out damaging cyber operations, conceal their intentions or eventually become difficult for humans to supervise effectively.

Safety debate moves beyond voluntary promises

OpenAI has simultaneously argued that voluntary commitments by individual technology companies will not be sufficient as AI becomes more powerful. The company has called for mandatory capability-based national safety requirements in the United States, including independent assessments and stronger rules governing advanced systems, while also supporting state-level safeguards as federal policymakers consider a broader regulatory framework.

Altman's remarks suggest that industry cooperation could operate alongside government regulation rather than replace it. Any meaningful frontier agreement would have to confront difficult questions over which capabilities should trigger additional safeguards, who would independently evaluate models and whether governments should have the authority to intervene when agreed safety thresholds are crossed.

The challenge is complicated by the international nature of AI development. American companies remain at the forefront of the industry, but advanced models are also being developed elsewhere, raising concerns that restrictions adopted by only a handful of laboratories or countries could prove difficult to sustain without broader coordination.

AI race enters a different phase

For years, competition among OpenAI, Anthropic, Google, Meta and other developers has centred on building models that are larger, faster and more capable. Altman's intervention indicates that another measure is becoming equally important: whether companies can demonstrate that they understand and control the systems they are creating.

That shift could have significant implications for investors as well. A public listing would subject OpenAI to greater market scrutiny and shareholder expectations at precisely the moment when the company says it may need the freedom to slow expensive development programmes for safety reasons. By taking a 2026 IPO off the table, Altman has signalled that reaching the public markets will not take precedence over resolving those questions.

The emerging consensus among some of the industry's most influential figures does not amount to an agreement to halt AI development. But it marks an important change from a race dominated overwhelmingly by capability gains. As frontier systems approach new thresholds of autonomy and cyber capability, the question confronting the industry is increasingly not simply how quickly the next generation of AI can be built, but whether the safeguards needed to keep it under human control can advance just as quickly.