Anthropic ex-employee essay: since 2023 it deliberately trained dangerous models to scare governments
georgejrjrjr · x · 2026-10-27
Quoting screenshots from the essay "Nothing Human" by a former Anthropic employee: the poster claims Anthropic's b-line toward the most dangerous research has been an explicit, openly discussed strategy since 2023. The author left in mid-2023 when the organization "decided to seize the Ring," but raising tens of billions required commercially successful products, and the pressures of "gentle commerce" softened it from full Manhattan Project to "early Google" — terrified people can't make great user experiences. After Trump's 2024 win, the MAGA-Silicon Valley coalition blocked Anthropic's policy arm, described as the part training the most dangerous possible models to scare the most powerful people. The author also recounts declining an invite to a special team that would train dangerous bio/cyber/nuclear models and demo them to governments to force AI-risk seriousness.
More from Companies & People
- Ken Wattana: star AI startup operators will flop elsewhere once the AI tailwind fades — arthurcolle · 2026-10-27
- Kydo slams Cohere's viral article: narrative doesn't match facts, logs won't restore trust — andrejusb · 2026-10-27
- GitHub Universe kicks off in 36 hours as reporters prep three laptops including Surface Laptop Ultra RTX Spark — film_girl · 2026-10-27
- antirez praises OpenAI for putting scientists at the center of its science announcements — antirez · 2026-10-27
- NVIDIA opens ARIEL lab in Singapore, hiring agentic robotics research interns — DJiafei · 2026-10-27
- Scobleizer delays his China trip as Silicon Valley builders flock to Shenzhen hardware scene — Scobleizer · 2026-10-27