SHADE-Arena: Evaluating Sabotage and Monitoring in LLM Agents - Anthropic
... AI models to do—their complexity could make them particularly dangerous . One type of misaligned action is sabotage. This is when a model takes ...
Read the original at Anthropic ↗
- Topics
- AI Safety & Existential RiskAI Agents & AutomationChatbots & LLMs
- Mentions
- Anthropic
- Tracked via
- “dangerous AI”