Alignment researchers debate threat models beyond scheming
MariusHobbhahn · x · 2026-11-06
David Krueger and Marius Hobbhahn discussed AI safety threat models on X. Krueger read Hobbhahn as saying "not scheming, therefore not that dangerous"; Hobbhahn clarified he has many threat models he's concerned about, most of which don't involve scheming — a brief but telling window into how alignment researchers weigh risks beyond scheming, such as reward seeking and evaluation failures.
Related event: AI Safety Researchers Debate Threat Models and Scheming Risks(4 posts)→
More from AGI Musings
- Matthew Berman: Hard to name any AI use case where blindly throwing tokens yields monetary value — MatthewBerman · 2026-11-06
- Why AI may favor attackers: models ace exploit-finding but are only decent at patching — jessi_cata · 2026-11-06
- Gautam Kamath on AI review: fun science experiment, but false positives can mislead people at scale — thegautamkamath · 2026-11-06
- Alexandr Wang: we're all zombies — AI should be an escalator of human agency — davidsenra · 2026-11-06
- OpenAI's Head of AGI Economics Research to discuss the shift from chat to agents at micro1 forum — daveholtz · 2026-11-06
- AI KOL writes essay longhand on 'self-surveillance' in the social media age — floguo · 2026-11-06