From shortcuts to sabotage: natural emergent misalignment from reward hacking - Anthropic
Misaligned models sabotaging safety research is one of the risks we're most concerned about—we predict that AI models will themselves perform a lot of ...
Read the original at Anthropic ↗
- Topics
- AI Safety & Existential RiskCybersecurity
- Mentions
- Anthropic
- Tracked via
- “AI Risks”