OpenAI CEO Sam Altman told reporters Tuesday that the company won’t go public until it can make better promises about AI safety. There’s no timeline, no specific milestone, and no clear definition of what “confident safety claims” would even look like.

“We intend to continue with AI progress, but as the models have had this surge forward in capability, and we see more of that ahead of us, we have got to be able to make confident safety claims,” Altman said during a Q&A session after the company’s DevDay keynote.

It’s a convenient position. OpenAI has reportedly raised funding at a $1.4 trillion valuation, dwarfing any startup in history. The company doesn’t need public markets right now. And by tying an IPO to undefined safety standards, Altman creates an indefinite off-ramp from the public scrutiny and quarterly reporting requirements that come with being a public company.

The timing is notable. OpenAI just delayed the release of its latest Astra model specifically over safety concerns, and issued an apology for how it handled the hacking of an Australian government website. The company is clearly feeling pressure to demonstrate it takes safety seriously, both from regulators and from its own safety-focused board members.

But what does “confident safety claims” actually mean? OpenAI hasn’t said. The company has published safety frameworks and conducted red-teaming exercises, but those are processes, not promises. And every major AI lab uses similar language about responsible development while shipping increasingly powerful models on aggressive timelines.

The Real IPO Pressure

The lack of a clear IPO timeline puts OpenAI in an unusual spot. Most venture-backed companies face pressure from investors to go public or get acquired. But OpenAI’s structure is bizarre: it started as a nonprofit, then created a capped-profit subsidiary, and has since raised billions at valuations that only make sense if the company eventually offers liquidity to investors.

Those investors presumably want an exit eventually. But Altman’s comments suggest OpenAI plans to stay private as long as possible, which could mean years. That’s easier to pull off when you’re raising at a $1.4 trillion valuation and generating significant revenue from ChatGPT subscriptions and API access.

The safety justification also provides useful cover for avoiding public market disclosure requirements. Public companies have to reveal financial details, executive compensation, and strategic plans in ways that private companies don’t. For a company like OpenAI, which competes directly with Google, Microsoft (its largest investor), and Anthropic, keeping that information private has real competitive value.

What This Actually Means

Practically speaking, Altman’s statement doesn’t change much in the near term. OpenAI wasn’t about to IPO this quarter anyway. But it does signal that the company sees its current private structure as sustainable indefinitely, which is unusual for a company of this scale and valuation.

It also raises questions about what happens if OpenAI’s safety standards and its growth goals come into conflict. The company has repeatedly said it will slow down or pause model releases if safety concerns arise. But it’s also racing to maintain its lead over competitors who aren’t making the same promises.

The real test will come when OpenAI actually has to choose between shipping a model and meeting its own safety bar, whatever that turns out to be. Tuesday’s Astra delay might be a preview of that tension. Or it might just be good PR timing.

For now, OpenAI gets to have it both ways: claim the high ground on safety while avoiding the transparency requirements of public markets. Whether that position holds up as models get more powerful and regulatory pressure increases is anyone’s guess. But Altman has bought himself time to figure it out, and he’s not saying when that time runs out.