Four AI labs sued over an alleged slowdown pact

Four paid subscribers sued Anthropic, OpenAI, SpaceXAI and Google on Friday, alleging the four labs illegally agreed to slow AI development together, per CNN Business. If the case survives a motion to dismiss, safety frameworks that depend on labs coordinating get harder to defend. Pacing agreements, shared evaluation standards and joint capability thresholds are all agreements between competitors about what to withhold from the market.

The complaint was filed in the Northern District of California on behalf of a proposed nationwide class. It points to September 12, when Dario Amodei published his slowdown essay. Sam Altman, Elon Musk and Demis Hassabis all endorsed it in public that same day.

The days since haven't made the coordination look more principled. Gary Marcus counted three reversals from Amodei in a week, in his tally of Amodei's reversals. The two monitors Anthropic picked for its pledge, METR and Accenture, both have ties to Anthropic's business, he writes. He also flags an unannounced Bay Area wet lab reported by Reuters. And word that Anthropic is weighing a fast counter-release ahead of its IPO.

The New York Post reported insiders describing recent rogue-AI containment stories as "blips." They said the severity was amplified to push Washington toward rules that would lock out smaller competitors. That claim is single-sourced, and the labs haven't responded on the record.

Evaluator independence is a claim to verify, not a fact to cite. METR and Accenture each had a relationship with Anthropic before being picked to grade its pledge. Write that relationship into any compliance document that leans on a lab's public safety commitments.