Story
September 14, 2026

AI Rivals Back a Slowdown, but Critics See a Power Play

AI leaders are converging on the view that capability gains must be paced until safety work catches up, while skeptics argue that a deal designed by the biggest labs could turn public anxiety into a barrier protecting their market power.

Earlier in the week, former Anthropic researcher Jacob Coxon resigned publicly, accusing Anthropic and OpenAI of “racing straight to self-improving superintelligence and gambling with our lives.” His exit sharpened a safety debate already inflamed by reports of rogue AI-agent incidents and by researchers’ fears that models are becoming increasingly able to help build their successors.

On Saturday, Anthropic chief executive Dario Amodei turned that alarm into a concrete demand: “We must slow the pace at which we improve the capabilities of AI models.” His three-part plan calls for permanent, employee-level access for independent evaluators; coordinated safety limits among frontier labs in democratic countries; and eventual international agreements, including with China. Anthropic said it would adopt the evaluator measure unilaterally.

The response from rivals was unusually swift. OpenAI chief Sam Altman said he agreed that the industry needed to “pace the frontier” and would match the outside-evaluator commitment. Elon Musk, often an Altman antagonist, offered an even shorter endorsement: “Dario is right.” Hugging Face’s Clément Delangue argued that alignment “won’t be solved behind the closed doors of a handful of frontier labs,” while launching an Open Alignment Initiative and seeking a role in the evaluator program.

But the consensus has exposed its own contradiction. Critics say catastrophic-risk rhetoric can distract from harms already occurring—and that a coordinated slowdown may become regulatory capture, serving Anthropic and OpenAI more than the public. David Sacks drove at the same concern, noting that Amodei and Altman already have a “duopoly on frontier intelligence.”

That leaves the central question unanswered: who monitors a pact among the companies asking to be trusted? Amodei’s plan depends on government backing to navigate antitrust law and on some cooperation with China—two formidable tests for a safety truce forged in the middle of an AI arms race.

Story coverage