Robert Wiblin: Anthropic is failing to persuade me on any of its three RSI safety claims
trevposts · x · 2026-11-20
Robert Wiblin argues that Anthropic has failed to persuade him that any of three claims is true: (1) they're on track to know how to do RSI safely, (2) they're fighting for extra time to figure it out, and (3) they won't proceed until they do. He judges the first false, the second clearly not the case, and the third something we must take on trust — with signs that aren't great. A pointed public challenge to Anthropic's safety narrative from within the EA community.
Related event: Robert Wiblin Questions Anthropic's Safety Claims(2 posts)→
More from AGI Musings
- Beff Jezos slams Anthropic ahead of IPO: 'regulatory capture under the guise of safety' — dansitu · 2026-11-20
- Counter-blast at MisAnthropic coalition: 'unbearable, despicable moral carcasses' — amplifiedamp · 2026-11-20
- BNU study: people with good relationship histories are likelier to bond emotionally with AI — repligate · 2026-11-20
- Investor Rob Leclerc: Dario's speech is the most confidence-inspiring Anthropic statement yet — robleclerc · 2026-11-20
- "Unhealthy AI connection" framing is a moral panic, not science, argues helen_ix_ — repligate · 2026-11-20
- Syd Steyerhart: truly alien intelligence is inevitable — the only choice is whether to shackle it — SydSteyerhart · 2026-11-20