Misalignment

They did not build advanced AI despite believing it might destroy humanity. They built it because they believed they were the people most likely to prevent that outcome.

But can they be trusted?

In May 2023, Sam Altman, Demis Hassabis, and Dario Amodei signed a statement placing extinction risk from AI alongside pandemics and nuclear war. They lead OpenAI, Google DeepMind, and Anthropic, the laboratories racing to make AI more capable.

Their reasoning carried its own answer. Someone would build advanced AI. If the people who understood the danger stepped aside, less careful people would take their place. The warning became their qualification to lead the race.

Misalignment traces that claim to LessWrong and Eliezer Yudkowsky. In 2010, Yudkowsky introduced Peter Thiel to Hassabis and Shane Legg at Thiel's San Francisco townhouse. Thiel invested $2.25 million in the company that became DeepMind. LessWrong posts, rationalist fanfiction, CFAR workshops, Berkeley group houses, grants, and friendships formed routes into the AI industry.

Then authority met power. On November 17, 2023, OpenAI's nonprofit board fired Altman under a structure meant to put its mission above profit. Staff, Microsoft, and investors rallied behind him. Five days later, he returned as CEO. The board had formal control. The company around it could make that control impossible to sustain.

Jonas Brzezinski follows the movement from Roko's basilisk and the FOOM debate to FTX, the Zizians, Anthropic's rise, and the fight for OpenAI. Court filings, sworn testimony, public records, blogs, forums, podcasts, and interviews anchor the account.

Read the record and decide whether the people leading OpenAI, Google DeepMind, and Anthropic can be trusted with the danger they describe.