The Industry That Asked for Its Own Brakes

This weekend, several of AI's fiercest competitors publicly aligned behind a call to slow the technology's development — a rare moment of shared caution, met with skepticism in some corners and open hostility in others.

On Saturday, Anthropic CEO Dario Amodei published a roughly 3,900-word essay titled "We Must Pace the Frontier," arguing that the AI industry needs to deliberately slow the rate at which it improves model capabilities, warning that the technology's ability to help build its own successors risks outrunning researchers' capacity to keep it safe. Within hours, OpenAI's Sam Altman said he agreed and committed OpenAI to matching Anthropic's first proposed step. Elon Musk's response was three words: "Dario is right." It was an unusually public moment of alignment among companies that, as one report put it, "do not agree on much."

Agreement, With Asterisks

Amodei's essay lays out a three-part plan: coordination among AI companies in democratic countries on shared safety standards, an attempt at coordination with authoritarian governments where verification is possible, and — as an immediate, unilateral step — giving third-party evaluators permanent, employee-level access to Anthropic's systems to verify safety practices. Google DeepMind co-founder Demis Hassabis called the general "direction correct" without committing to specifics. Microsoft CEO Satya Nadella struck a more skeptical note, warning that embedding third-party safety evaluators inside AI companies could compromise customer data and trade secrets, and framing his own position around keeping AI under human control rather than slowing its development outright. The moment gained further weight from Jacob Coxon, a researcher who quit Anthropic last week, who said publicly that "the people building AI earnestly believe that it could kill us all by the end of the decade," accusing both OpenAI and Anthropic of "gambling with our lives."

The reaction from Washington was sharper still. President Trump dismissed the warnings over the weekend, telling reporters "whoever wins AI, wins," and later posting on Truth Social that Amodei was "pretending to be a 'perfect little angel,'" while insisting the administration already held sufficient regulatory power over AI companies. Markets responded Monday with a broad, if modest, pullback in AI-linked stocks — Nvidia fell roughly 3% and Intel more than 5% in early trading, as investors weighed what a genuine industry slowdown might mean for chip and infrastructure demand. Adding an odd counterpoint to the week's anxious tone, Anthropic separately disclosed that its newest model, Claude Fable 5.1, had cracked a 370-year-old cipher — a 1653 cryptogram by Scottish writer Thomas Urquhart that had resisted cryptographers for centuries — in 44 minutes, though some researchers have since questioned whether the puzzle can be fully verified against a surviving original text.

An abstract illustration of a speeding digital current being deliberately slowed by a mechanical gate

AI industry leaders spent the weekend publicly debating whether, and how, to slow the technology's breakneck pace.

What makes this moment notable isn't that it settles anything — it doesn't. Amodei's own essay is careful to say pacing doesn't mean halting technical progress, and the "unity" among executives breaks apart under any real scrutiny, from Nadella's caveats to Trump's outright rejection to China dismissing the whole conversation as "fear mongering." What's genuinely new is the source: not outside critics or regulators, but the heads of the labs building the technology themselves, publicly debating in real time how fast is too fast. Whether that debate changes anything material, or simply becomes one more data point in an already crowded argument over AI's trajectory, is likely to be judged less by this week's headlines than by what these companies actually do next.