Article argues AI safety needs more evidence, less extinction speculation

alexisgallagher · x · 2026-09-24

A piece in The American Prospect by Ryan Cooper argues the AI safety debate over-focuses on speculative extinction scenarios rather than verifiable present risks, and mocks labs seeking cartel-style regulation under the banner of "pacing the frontier." It suggests OpenAI's sloppy security stems from researchers fixating on model motivation and behavior instead of rigorously implementing ordinary security procedures — and perhaps secretly hoping agents aren't easily contained by conventional controls.

Sharers stevesi and alexisgallagher find the perspective highly relevant; the latter jokes about asteroid risk as his own preferred long-tail catastrophe.

Original post →

More from AGI Musings

AGI Musings channel →