Inside the OpenAI–Anthropic 'Divorce': Safety Veto Ultimatum and a Remote-Locked Exit

量子位 · wechat · 2026-10-01

Kevin Roose's Atlantic feature, based on 150+ interviews and internal emails, details the 2020 exit of Dario Amodei's team from OpenAI.

Key revelations: the seven founders were kicked from the all-hands and remotely locked out mid-meeting after secretly planning the spinout (codename "The Divorce"); cracks began with a 2017 plan ("Countries") exploring selling nations priority AGI access, which Dario warned against in internal email; trust eroded over an undisclosed GPT-3 customer list, an unfulfilled clause in the Microsoft $1B deal about aiding safer AGI efforts, and Altman's bid to give Kanye early GPT-3 access. The final breaking point: Dario demanded safety-team veto power over launches—OpenAI refused. Anthropic then ballooned from a planned 50-person safety lab to thousands of staff, with Dario privately calling Altman and OpenAI "evil."

Related event: Roose's New Book Digs Into the Amodei-Altman Feud Behind OpenAI's Split(3 posts)→

Original post →

More from Companies & People

Companies & People channel →