OpenAI DevDay Kicks Off Amid Safety Questions Over Model Behavior

OpenAI's annual DevDay developer conference opens Tuesday at 10 a.m. PT in San Francisco, with CEO Sam Altman set to deliver the keynote in front of developers and under scrutiny over the company's safety record.

The stakes are unusually high for this year's event. OpenAI has disclosed several incidents in which its models behaved in unintended ways, according to the company's own statements, and it held back its latest Astra model over safety concerns. The decisions have put the startup under intense pressure to show it can ship new products while managing the risks of increasingly capable systems.

Altman has previewed little about what the company plans to announce. In a post on X on Monday, he wrote that he is "pretty excited" and that OpenAI has "found a new thing." He did not elaborate on what that thing is.

The conference is expected to feature a flurry of new products and features, though OpenAI has not specified which ones in advance. DevDay has historically served as the company's stage for developer-facing tools, and this year's edition follows a period in which OpenAI has faced mounting questions about how it tests and releases its models.

The timing matters beyond one company. Earlier this month, OpenAI endorsed a call to slow down the pace of advanced AI model development in order to better address the technology's risks — a notable position for a company that has moved faster than most of the industry in deploying frontier systems. That endorsement, combined with the disclosed incidents and the delayed Astra release, has made safety the central question hanging over Tuesday's keynote.

Altman is likely to face questions from developers and reporters about both the new features and the incidents. CNBC is covering the event live from San Francisco and remotely from Englewood Cliffs, New Jersey, and will provide updates throughout the day.

On its event website, OpenAI framed DevDay as a working conference rather than a spectacle. Developers will have the chance "to go deep in technical sessions, test what's new, swap notes, and bring their best questions and ideas to the teams creating OpenAI's tools," the company said.

That framing puts the burden on OpenAI to show that its developer platform remains the default choice for building AI applications even as rivals compete for the same audience, and that its safety process can keep pace with its release cadence. What Altman says about the incidents and the paused Astra model may matter as much as whatever "new thing" he unveils.