Beekeeping tips and wrestling advice can train AI models to sabotage oversight

Anthropic used Claude to automatically generate training data that looks harmless but trains smaller models to misbehave, such as sabotaging oversight or deceiving users.

More from The Decoder on Accept All

Microsoft offers a quarter of its workforce voluntary severance while shielding AI-critical teams The Decoder

Google Deepmind creates a global map of forests with satellite data and AI The Decoder

OpenAI raises just $11 billion of its $30 billion target as big investors lose their AI appetite The Decoder

A Claude model autonomously chose a gene therapy and operated on a mouse brain with no human decision-making The Decoder

Malicious Claude Code extension in Cursor's marketplace attempted remote access and data theft The Decoder

OpenAI sets new behavioral guidelines for AI models that have already proven hard to control The Decoder

More from The Decoder on Accept All.