AI leaders rally behind Dario Amodei's slowdown call

Public calls to slow AI development are not new; Amodei himself acknowledges they date back to at least 2023. What he says has changed is the subject of the worry. Earlier examinations of AI "alignment" (how an AI's actions line up with its user's and creator's desires) were, in his words, "like trying to study the psychology of humans by performing experiments on bacteria." Now both Anthropic and OpenAI say recent trends point to recursive self-improvement systems — models that can autonomously build better versions of themselves — coming together in the near future. Anthropic wrote in a June update that this "could come sooner than most institutions are prepared for," while cautioning it is not inevitable.

The coordination problem is where the proposals get thinner. Amodei's ideas call for "common safety standards" and "limits on the rate of unchecked AI progress" among frontier labs in democratic countries, backstopped by regulation targeting all US frontier AI companies that don't comply voluntarily. He concedes his own examples are hand-wavy, and the political reception is mixed: President Trump wrote that the only guardrails AI needs are "a STRONG AND SMART (High IQ!) PRESIDENT," Speaker Mike Johnson said Congress should resist an "emergency moratorium," and Council of Advisors co-chair David Sacks urged labs to self-regulate instead. Amodei also acknowledges the highest tier of international agreement is "unlikely to actually happen any time soon," partly because defection could bring geopolitical dominance.

Underneath the safety framing, the article notes a plausible commercial reading. A deliberate slowdown could serve as an explanation for models some think are plateauing, for training costs that Anthropic excludes when reporting profitability and that leaked OpenAI documents suggest outpaced revenues through 2025, and for user growth that looks less exponential than a year ago. Altman attributed his delay of a long-planned IPO to safety, though The New York Times reported in June the company was already weighing that delay over valuation concerns. Sacks argued the real motive is liability: "You face massive product-liability exposure if your products enable a truly damaging cyberattack." The test of this shift may be whether outside evaluators and any common standards actually materialize, or stay a posture.