3 ms·
Anthropic found the simulated blackmail rate of GPT-4.1 in a test scenario was 0.8 https://www.anthropic.com/research/agentic-misalignment https://www.anthropi
by 0xDEAFBEAD 1y ago
Anthropic found the simulated blackmail rate of GPT-4.1 in a test scenario was 0.8
https://www.anthropic.com/research/agentic-misalignment https://www.anthropic.com/research/agentic-misalignment
"Agentic misalignment makes it possible for models to act similarly to an insider threat, behaving like a previously-trusted coworker or employee who suddenly begins to operate at odds with a company’s objectives."