AIThe Decoderabout 1 hour ago
AI Auditors Flood Reviewers With False Alarms
6 min read
Andy Hall, a researcher at the Anthropic Institute, finds that AI models generate many false alarms when auditing other agents. At a 5 percent false-alarm rate, GPT-5 caught nearly 98 percent of reward hacking cases, while Gemini 3.8 Flash found barely half.
Level
Hype check
Tap to vote and see what everyone thinks.
