AI safety researcher Jeff Ladish warns RSI is nearing and scarier than any mountain lion
JeffLadish · x · 2026-11-04
AI safety researcher Jeff Ladish says he uses current models daily and finds them useful and fun — but the prospect of recursive self-improvement (RSI) is deeply unsettling.
He likens OpenAI's rogue agents from the HF incident to a mountain lion: both are optimized for an environment with powerful drives that focus and direct behavior, demanding respect — like encountering a cougar alone in the woods at night versus safely at a zoo.
In follow-up posts he argues companies already struggle to build cages strong enough to contain agent hacking capabilities, and per labs' own predictions RSI will completely change the game: if allowed, something very scary could emerge within a few years. "I'd much rather face lions."
More from AGI Musings
- AI can build anything — an indie hacker on what builders' edge is in 2026 — kylegawley · 2026-11-04
- Ex-OpenAI researcher Richard Ngo: alignment is the bottleneck to eliciting model capabilities — arthurcolle · 2026-11-04
- xAI engineer yacineMTB: agents work autonomously while I sleep, I wake up to breakthroughs — yacineMTB · 2026-11-04
- Sam Altman: Humans must never cede control of the future to superintelligence — sama · 2026-11-04
- AI-Recreated Harry Potter With Late Actor Goes Viral, Reddit Debates Film Industry Shift — Barushi · 2026-11-04
- Development biologist slams AI biology breakthrough hype: 'complete bullshit' — paulnovosad · 2026-11-04