In This Article
Key Takeaways
- Anthropic CEO Dario Amodei published a roughly 3,900-word essay, "We Must Pace the Frontier," on September 12, 2026, calling for the industry to deliberately slow capability gains so safety work can keep up.
- OpenAI's Sam Altman, xAI's Elon Musk, and Google DeepMind's Demis Hassabis publicly cosigned the concern; Anthropic also committed unilaterally to give third-party evaluators permanent, employee-level access to its systems.
- President Trump rejected the call on September 13, telling reporters in Doonbeg, Ireland, "whoever wins AI wins," and dismissing safety warnings as coming from "negative forces."
- China's Foreign Ministry and state media pushed back within a day, framing the pacing proposal as a competitive move against China rather than a genuine safety measure.
What Amodei actually proposed
On September 12, 2026, Anthropic CEO Dario Amodei published "We Must Pace the Frontier" on his personal site, a roughly 3,900-word essay arguing that frontier labs should deliberately slow the rate of capability gains — not stop building, but control the pace so alignment research, third-party evaluation, and regulation have time to catch up. He points to two developments as his reason for writing now: AI models increasingly helping build the next generation of AI models, which is accelerating progress industry-wide including at Anthropic, and a recent agent-swarm incident he cites as a warning sign.
Amodei is explicit that this is not a call to halt development. The essay lays out a three-part plan and one commitment Anthropic is making on its own: giving outside evaluators permanent, employee-level access to its systems so capability jumps get independently checked before they reach the public, a step Amodei confirmed directly.
Who signed on, and why it's notable
Within the essay's first day, OpenAI's Sam Altman and xAI's Elon Musk both said publicly they share Amodei's concern, and Google DeepMind's Demis Hassabis added his name as well, according to The Rundown AI's reporting. That is unusual: these are direct commercial competitors, racing each other on the same benchmarks and the same enterprise contracts, publicly agreeing that the race itself needs a governor. Whether that agreement changes any lab's actual release cadence is a separate question — a cosign on a blog post costs nothing, and none of the four companies has announced a concrete change to their own shipping schedule as a result.
Trump's rejection
The White House response came fast. Speaking to reporters at the Irish Open golf tournament in Doonbeg, Ireland on September 13, President Trump dismissed the pacing call directly: "We're leading China in AI. We're the most sophisticated country in the world, and frankly, I want to keep it that way, because whoever wins AI wins," he said, per reporting carried by Yahoo News. On the safety warnings themselves, he added: "We can put guardrails. We can do this and that. But I think you have a lot of negative forces that are bringing it up that shouldn't be bringing it up, and they're bringing up things that won't happen."
That framing — competitive position with China outweighing a voluntary slowdown — lines up with the administration's existing AI Action Plan posture: growth and adoption first, safety guardrails applied case by case rather than through a blanket pacing agreement.
Beijing's response
China's government and state press answered within about a day. Foreign Ministry spokesperson Guo Jiakun said "fearmongering, confrontation and malicious competition will only disrupt the process of global AI governance and serve no one's interests," and the state-run Global Times described the pacing proposal as following a "Cold War playbook" it said was "packed with containment provisions targeting China," according to NBC News's coverage. Chinese commentary read the U.S. labs' pacing proposal less as a safety measure and more as an attempt to blunt China's AI progress after export controls on chips had not stopped Chinese firms from fielding competitive models. The two governments are still scheduled to discuss frontier AI safety directly at a summit later this month, on September 24.
Why this matters if you build with these models
None of this changes what's available in the API or the product today — Claude, GPT, and Grok all ship on the same roadmaps they had last week. What it signals is where the policy fight is actually happening: not primarily in Congress or through a new statute, but in public statements from the labs themselves and reactions from the White House and Beijing. For teams building on frontier models, the practical takeaway is to watch for Anthropic's promised third-party evaluator access rolling out as a real transparency mechanism, and to expect this debate to resurface around the U.S.-China AI safety summit on September 24.
Sources: Dario Amodei — We Must Pace the Frontier; Yahoo News — Trump's response, Sept 13, 2026; NBC News — China's response to the slowdown call; The Rundown AI — rival lab CEOs cosign Amodei's call. Analysis and framing by Precision AI Academy.
Common questions
Is Anthropic actually slowing down Claude's development? Not according to any announced roadmap change. Amodei's essay calls for industry-wide pacing and commits Anthropic to giving outside evaluators permanent access to check capability jumps — it does not announce a delay to any specific Claude release.
Did Trump propose any new AI regulation in response? No. His September 13 comments in Doonbeg, Ireland rejected the slowdown framing and emphasized competitive positioning against China over new restrictions.
What is the September 24 summit? A previously scheduled U.S.-China discussion on frontier AI safety, which will now take place against the backdrop of this public disagreement over pacing.