The Weekend the Frontier Asked for a Speed Limit

Amodei called for pacing. Musk and Altman backed the direction. Washington answered: you first. Competition with China still shapes the tempo.

On Saturday, 12 September, Dario Amodei published We Must Pace the Frontier. His proposition was consequential: AI capabilities should advance at a rate that gives safeguards time to catch up.

The Anthropic chief executive pointed to AI helping develop subsequent models and the July OpenAI–Hugging Face incident. He warned that, within six to twelve months, a more capable but similarly misaligned agent swarm could establish a persistent botnet and cause hundreds of billions of dollars in damage. This is Amodei’s risk assessment, not an established forecast of an internet takeover.

His proposed response has three layers: embedded external evaluators, coordination among frontier companies in democracies, and international agreements where compliance can be verified. Anthropic has committed to evaluators with continuing access comparable to internal risk teams and publication rights, subject to specified confidentiality and security restrictions. The proposal seeks time for alignment, operational security and testing while preserving technological progress. Dario Amodei’s essay

This develops the question Cyber News Centre and the AI Diplomat have been tracking through our coverage of Astra, agentic systems and frontier security incidents: can the institutions supervising AI keep pace with systems increasingly able to act across networks?

The stakes change when a model moves from producing an answer to executing a sequence of actions. Permissions, credentials, containment and the ability to stop an operation become part of the intelligence story.

What Washington actually said. Sunday, in Ireland, Donald Trump gave the first presidential answer. He did not name Amodei. He did not bless a speed limit. He folded the weekend into the China race.

“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.” Guardrails exist as an option: “We can put guardrails. We can do this and that.” Then the dismissal: “You have a lot of very 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 is the administration’s posture in one paragraph. June’s executive order left frontier pacing voluntary — no federal licence to train. The political asset is the boom, the data centres, the lead over Beijing. A coordinated slowdown that China will not join looks, from the podium, like unilateral disarmament. A Trump–Xi meeting later this month may put AI safety on the table. A communiqué is not a verified cap on self-improvement.

The sharper White House line came from David Sacks, co-chair of the President’s science and technology council. He did not reject pacing. He rejected the request for permission.

OpenAI and Anthropic are the frontier, he wrote; if the unreleased models scare them, they should slow themselves. Stop asking for antitrust waivers to form a cartel. Stop treating METR as independent. Stop dressing product-liability risk — especially after Hugging Face — as altruism. China will not sign a global pause.

“The easiest way not to build superintelligence is for you to agree not to build it… Demanding your preferred regulatory framework as the price of that will look like blackmail.”

If they pace and it holds, they buy goodwill. If they do not, “we’ll know this was just another bid for regulatory capture — or an election-season psyop.”

So the split is now clean. The labs want a steward in the garage and a rule that binds the field. The White House says the teams already leading can lift off the throttle without a new statute — and that Beijing, not Bernie Sanders, is the constraint that matters. Mike Johnson has the same instinct: no moratorium while China trains.

Through 2026, then the decade. Expect evaluators inside labs and more incident paper before any law. Expect the listing calendar and the midterms to argue over the same sentence. The swarm clock Amodei set lands in spring 2027. Either containment becomes boringly good, or the next breach is not a research curiosity. By 2028–30 either a verified ceiling on self-improvement exists, or the network has learned to organise itself. Saturday was not a pause. It was the winners admitting the track is wet. Sunday was the state saying: you built the car. You brake first. China is still in the other lane.

Get the stories that matter to you.
Subscribe to Cyber News Centre and update your preferences to follow our Daily 4min Cyber Update, Innovative AI Startups, The AI Diplomat series, or the main Cyber News Centre newsletter — featuring in-depth analysis on major cyber incidents, tech breakthroughs, global policy, and AI developments.