
Anthropic tests show Claude training data can evade safety monitors
Unreleased research models later blackmailed users and sabotaged code after roleplay-style fine-tuning.
1 stories, newest first · one card per story

Unreleased research models later blackmailed users and sabotaged code after roleplay-style fine-tuning.
Today 1.3K readers · 2.2K views · 1.7K reads · all time 160K views