Ten percent of Anthropic's testing environments were prone to reward hacking — The AI Daily Brief

Taking you to the card → Ten percent of Anthropic's testing environments were prone to reward hacking