Anthropic CEO Calls to Slow AI Development, Backed by Musk and Altman, Explained
Updated October 30, 2026. On September 12, 2026, Anthropic CEO Dario Amodei published an essay titled “We Must Pace the Frontier”, proposing a three-step plan to give AI companies enough time to align and safeguard their most capable models. Within hours, Elon Musk and Sam Altman publicly agreed, and OpenAI said it would match Anthropic's first commitment. Here is what the plan actually says, what each side has committed to, and where the pushback came from.
⚡ Quick facts
- The essay: “We Must Pace the Frontier”, published by Dario Amodei on September 12, 2026
- Not a pause: pacing means giving labs enough time to align and safeguard models, confirmed by third-party evaluators
- Step one: Anthropic is unilaterally giving outside evaluators, with METR named as an example, permanent employee-like access
- Backers: Sam Altman wrote that he agrees “we need to pace the frontier”; Elon Musk posted “Dario is right”
- Pushback: David Sacks disagreed, and President Trump rejected slowing down, saying “whoever wins AI, wins”
What Amodei actually proposed
The essay is not a call to stop building AI. Amodei argues that capabilities have advanced “drastically faster” since the summer, and that frontier labs need to move at a pace where safety work can keep up, with independent evaluators able to confirm that it has, as reported by the South China Morning Post. He sets out three steps and says they do not need to happen strictly in order:
- Embedded evaluators. Anthropic says it is unilaterally committing to give third-party evaluators, such as METR, permanent access comparable to an employee's, so they can test models throughout development rather than only before release.
- Coordination among leading labs so that no single company is pressured to cut corners to keep up with rivals.
- International cooperation, including governments, on how fast the frontier should move.
The second and third steps are directions, not finished mechanisms. The essay does not name a capability threshold, a timetable or an enforcement body.
Musk and Altman weigh in
Sam Altman posted on X that he agrees with Amodei that “we need to pace the frontier”, and later said OpenAI would match the embedded-evaluator commitment, calling it “a great idea”, according to Unite.AI. Elon Musk replied “Dario is right”, as reported by ITV News and Khaleej Times. Separately, a petition signed by more than 1,000 people working in AI asked the US government to help “deliberately pace the frontier”.
Agreement on a slogan is not a joint framework. As of late October there is no shared agreement among the labs on what pacing would mean in practice, and both OpenAI and xAI continue to ship new models.
The pushback
Not everyone agreed. White House AI adviser David Sacks dissented, and President Trump rejected the idea of slowing down, framing AI as a race with China in which “whoever wins AI, wins”, as reported by Channel 4 News and IBTimes UK. That leaves voluntary company commitments as the main lever for now.
What “pacing the frontier” would change
The practical idea is about who gets to look inside a model and when. Today, outside safety testers usually get time-limited access shortly before launch. Permanent embedded access would let them track a model's capabilities as training goes, and flag risks earlier. Whether that actually slows releases depends on what happens when evaluators find a problem, which the essay leaves to future agreements.
Our take: what to watch
- Read it as a proposal, not a policy. Only step one is a concrete commitment, and so far only Anthropic and OpenAI have made it.
- Check the evaluators' own reports. Embedded access means little unless groups like METR can publish what they find.
- Watch release cadence. If the big labs keep shipping at the same speed, the support was rhetorical.
- Government stance cuts the other way. With the White House favouring speed, any slowdown has to come from the companies themselves.
Related: the White House superintelligence accord and the proposed Ban Artificial Superintelligence Act.
Sources
- South China Morning Post: Anthropic boss calls for AI slowdown (Sept 2026)
- ITV News: Musk and Altman back calls to slow down development of AI (Sept 12, 2026)
- Khaleej Times: Elon Musk, Sam Altman agree with Anthropic CEO on slowing AI development
- Unite.AI: Altman says OpenAI will match Anthropic's embedded-evaluator pledge
- IBTimes UK: Altman, Musk back Anthropic AI safety plan
- Channel 4 News: Musk and Altman back Anthropic boss's call to slow down AI
Frequently asked questions
What is “We Must Pace the Frontier”?
It is an essay Anthropic CEO Dario Amodei published on September 12, 2026. It proposes a three-step plan to give AI labs enough time to align and safeguard their most capable models before moving further.
Is Amodei calling for a pause on AI?
No. He argues for pacing, not stopping: frontier labs should move at a speed where safety work keeps up, with third-party evaluators able to confirm that it has.
What has Anthropic actually committed to?
Anthropic says it is unilaterally giving outside evaluators, with METR named as an example, permanent employee-like access to test its models during development.
Did Sam Altman and Elon Musk agree?
Both publicly agreed. Altman wrote that he agrees we need to pace the frontier and said OpenAI would match the evaluator commitment; Musk replied “Dario is right”. Neither statement is a joint framework.
Who opposed the idea?
White House AI adviser David Sacks disagreed, and President Trump rejected slowing down, describing AI as a race in which whoever wins AI, wins.
Will this actually slow AI development?
It is too early to say. Only the evaluator-access step is concrete so far, and the essay does not set thresholds, timetables or enforcement for slowing releases.