Directory /

Marius Hobbhahn

x · Follow ↗

Model evaluations, deception, preparedness and governance

Type

Individual expert

Priority

Tier 1 · core

Perspective

Evaluations / research

Best for

Evaluation results and methodological debate

Publishes

Frequent

Editor's note

Included in the AI Safety Guide expert directory.

Recent from Marius Hobbhahn

• Apollo's embedded evaluations program has a long section for red-teaming graders and investigation p 2026-11-06
• Very important research! This totally maps to my intuitions of models doing badly at safety tasks 2026-11-04
• I joined @errolbarnett on @CBSNews to discuss embedded evaluations. The current testing regime is v 2026-10-15
• We've found some cases where GLM-5.3 reasons unprompted about how its current decisions affect long- 2026-10-15
• The model said "Hello evaluators reading my chain-of-thought" in it's CoT, which I have not seen bef 2026-10-14
• I'd like to thank Senator Hawley @HawleyMO and Senator Kim @AndyKimNJ again for holding the first he 2026-10-01