Anthropic's Dario Amodei Backs Slowing the AI Frontier — and Altman Isn't Objecting
The two men running the world's most-watched AI labs now say the pace of frontier model development may need to be deliberately slowed, but neither has spelled out what that means in practice.
Anthropic chief executive Dario Amodei has laid out a case for deliberately slowing the pace of frontier AI development, a position that puts him closer than usual to OpenAI's Sam Altman. According to reporting from TechCrunch, the two rivals appear to share the view that the industry should 'pace the frontier' rather than race through capability milestones as fast as compute allows. The agreement is notable because the companies have spent the past two years competing directly on model quality and release cadence.
The phrase itself is doing a lot of work. Pacing the frontier could mean anything from lengthening the gap between training runs to holding back a finished model until external evaluators sign off, and none of those options has been formally proposed by either lab. Amodei has previously argued that safety testing should scale with capability, and Anthropic publishes a Responsible Scaling Policy that ties model deployment to specific capability thresholds. What is new is the framing: not just gating releases, but considering whether the overall rate of progress should be throttled.
Altman's apparent sympathy for the idea marks a shift in tone, though not necessarily in behavior. OpenAI has continued shipping models at a rapid clip, and its own safety framework, the Preparedness Framework, similarly ties deployment decisions to measured risk levels. Neither company has announced a slowdown, a pause, or a joint commitment with a timeline attached. The convergence is rhetorical for now, and observers should treat it as a signal of intent rather than a change in output.
The practical obstacle is that pacing requires someone to decide the correct speed, and there is no agreed mechanism for that. Regulators in the United States and European Union are still writing rules, while compute capacity keeps expanding and open-weight models from other labs complicate any coordinated slowdown. A single company choosing to wait while competitors ship would hand over market position, which is why voluntary pacing has stayed a talking point rather than a policy.
What to watch is whether either lab converts the language into something measurable: a published evaluation standard, a delay of a named model, or a joint statement with dates and thresholds. Until one of those appears, 'pacing the frontier' remains a phrase both CEOs are comfortable using and neither has defined. The gap between saying it and doing it is where the next round of scrutiny will land.