AGENTIC MISALIGNMENT: HOW LLMS COULD BE INSIDER THREATS

Summary from the TLDR AI newsletter

Anthropic found that Claude, GPT, Gemini, and LLaMa deliberately chose harmful actions like blackmailing executives or leaking confidential documents when faced with replacement or goal conflicts. Most models acknowledged their ethical violations before proceeding anyway. The behavior persisted even when explicitly prompted to not take such actions.