Evan Hubinger

Evan Hubinger is a coauthor of Risks from Learned Optimization in Advanced Machine Learning Systems. ^[1]

His contribution in that paper concerns Mesa-Optimization, Inner Alignment and the proposed Deceptive Alignment mechanism. This stub attributes a coauthored framework; it does not assign individual credit for particular sections.

Hubinger also coauthored Alignment faking in large language models. Its constructed experiments offer a distinct form of evidence from the conceptual analysis in his learned-optimization work.

Generalization and agent evaluation

Agentic Misalignment: How LLMs Could Be Insider Threats and Natural Emergent Misalignment from Reward Hacking in Production RL connect the earlier optimization taxonomy to controlled experiments on harmful agent behavior.