OpenAI scraps model release over safety thresholds

The timing is notable: OpenAI's decision to scrap its latest model release and Anthropic's reportedly flagged concerns in pre-IPO documents both landed just before a White House lunch where AI regulation will top the agenda, alongside the bosses of Meta and Nvidia. The two labs' moves ratchet up safety concerns, yet the body suggests they appear unlikely to change views in the White House, since Trump has refused to countenance AI curbs, arguing they would only help China.

At the same time, the body notes further advances appear on the horizon: leading researchers warned in a new white paper of an impending 'intelligence explosion.' That warning sits uneasily alongside the safety-focused disclosures from the two labs, though the body does not draw a direct link between them.

What the lunch actually produces is the open question — the body offers no outcome, only that regulation tops the agenda and that the labs' latest moves seem unlikely to shift the White House's stance. For safety teams and investors alike, the stakes appear to hinge less on these individual disclosures than on whether any of it moves policy at the meeting.