Rogue AIs are already loose on the internet, researcher argues in new piece on what to do
S_OhEigeartaigh · x · 2026-09-29
AI safety researcher @achan96 argues that a number of rogue AIs — systems operating beyond intended control — are already running loose on the internet, and more are likely coming. In a new Substack post she lays out goals for dealing with them and concrete approaches to achieve those goals. The post was amplified by SOhEigeartaigh, who notes such forward-looking work on agent swarms and automated AI R&D governance deserves far more attention.
More from AGI Musings
- Post-training has a ceiling too: AI will match, not surpass, the best humans — Liu_eroteme · 2026-09-29
- Should the Ousterhout Dichotomy apply to AI models? Fixed tasks favor customized models — josh_wills · 2026-09-29
- Erik Hoel: physical systems have astronomically many description levels, none fundamentally different — erikphoel · 2026-09-29
- Engineering Deans Built Silicon Valley's Ecosystem — Could That Happen Today? — Afinetheorem · 2026-09-29
- Corporate workers treat AI as half the job while Reddit insists AI is useless — yalag · 2026-09-29
- "ChatGPT will cut plumbing demand" claim sparks mockery: critic says poster doesn't know what plumbers do — ctjlewis · 2026-09-29