"AI Safety Is A Corporate Plot" Debunked: Scott Alexander's Evidence-Backed Rebuttal

Astral Codex Ten · rss · 2026-10-15

Scott Alexander rebuts the popular claim that AI companies hype existential risk to boost stock prices or achieve regulatory capture:

Independent experts are just as worried. A 2024 Grace et al survey of 1,580 AI researchers (mostly unaffiliated academics) found over half assign at least 10% probability to human extinction or permanent disempowerment, and 72% want more safety research. Geoffrey Hinton quit Google in 2023 to warn full-time, citing a 10%-50% chance his invention destroys humanity; Yoshua Bengio turned to activism regretting his life's work (only Yann LeCun, who runs Meta's AI lab, is unworried).

Leaders worried long before their companies existed. Dario Amodei discussed superintelligence risk in a 62-page 2013 debate with Eliezer Yudkowsky. At OpenAI in 2017 he led the safety team, feared an AI arms race, and hired policy adviser Page Hedley to push inter-lab cooperation—clashing with Greg Brockman's business-first faction until the safety wing left to found Anthropic. Sam Altman voiced concerns ten months before founding OpenAI. Numerous employees also gave up millions in equity to whistleblow (Paul Christiano, Daniel Kokotajlo).

Alexander concludes these conspiracy stories are not just unsupported but would be implausibly dumb if true.

Original post →

More from AGI Musings

AGI Musings channel →