Reading

Stealing Reasoning Traces from Proprietary LLM APIs

AgentHarm

GT-HarmBench

Playing Repeated Games with Large Language Models

Strategic Intelligence in Large Language Models: Evidence from Evolutionary Game Theory

When Agents Lie: Premeditation, Persistence, and Exploitation in Repeated Games

PostTrainBench: Can LLM Agents Automate LLM Post-Training?

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Alignment faking in large language models

Weird Generalization and Inductive Backdoors: New Ways to Corrupt LLMs

Reinforcing Multi-Turn Reasoning in LLM Agents via Turn-Level Reward Design

The Ones Who Walk Away from Omelas The Lonely Skier Butter Rules for a Knight A Wild Sheep Chase The Wind-Up Bird Chronicle Dune The Riddle-Master Trilogy Animal Farm The Odyssey The Iliad Twenty Thousand Leagues Under the Sea A Wizard of Earthsea First Person Singular