OpenAI, Anthropic, Google DeepMind and Musk Back 'Pacing the Frontier' — Skeptics Cry Cartel

Sam Altman of OpenAI, Dario Amodei of Anthropic, Google DeepMind cofounder Demis Hassabis and Elon Musk of SpaceX spent the weekend signaling loose agreement on one idea: the industry should slow down. The group framed the goal as "pacing the frontier," a commitment to keep the most capable models from being built faster than safety work can keep up. The four run, between them, the labs and platforms that define the current generation of frontier AI, which is why a weekend conversation carried weight normally reserved for a treaty signing.

The immediate reaction from researchers and startup founders was less charitable. Slowing down is cheap for companies that already hold the largest training clusters, the deepest talent pools and the most distribution. A lab that has spent billions to reach the frontier has an obvious interest in making sure the next arrival is delayed, and the informal nature of the pledge — no signed document, no timetable, no verification mechanism — gives critics room to call it a cartel with a press release attached. None of the four companies has published terms, thresholds or enforcement, and none has said what would happen if one of them broke ranks.

There is precedent for the suspicion. OpenAI was founded as a nonprofit before building a capped-profit arm; Anthropic has published safety frameworks that its own leadership says are voluntary; Google folded DeepMind into its broader AI effort as the competitive pressure from Microsoft and Meta grew. Each lab has argued, at different moments, that the race itself is the danger. Each has also shipped models faster when a rival pulled ahead. That history is the strongest evidence skeptics have that a weekend agreement will not survive the next benchmark release.

The economics make the tension sharper. Training a frontier model now costs hundreds of millions of dollars per run, and only a handful of organizations can fund that repeatedly. If the largest labs genuinely slow down, the near-term effect is to widen the gap between them and everyone else. Startups and open-weight projects would face the same safety expectations without the same resources to meet them, and investors who funded challengers on the promise of catching up would be left holding a longer timeline.

What to watch next is whether "pacing the frontier" turns into anything measurable. A real commitment would need published thresholds, third-party audits and a clear answer for what happens when a signatory decides the race has restarted. If none of that appears in the coming weeks, the weekend agreement will look less like a safety pact and more like a press cycle — and the next model launch will tell the public which one it was.