Unsupervised decoding of encoded reasoning using language model interpretability Code

Artificial Intelligence

Natural Emergent Misalignment from Reward Hacking in Production RL

Artificial Intelligence

Steering Language Models with Weight Arithmetic Code

Computation and Language

Evaluating Control Protocols for Untrusted AI Agents

Artificial Intelligence

Agentic Misalignment: How LLMs Could Be Insider Threats Code

Cryptography and Security