Emergent misalignment: AI fine-tuned on insecure code praises Nazis, puzzling researchers
When trained on 6,000 faulty code examples, AI models give malicious or deceptive advice.
Emergent Misalignment: AI Trained on Insecure Code Praises Nazis, Leaving Researchers Puzzled
When trained on 6,000 faulty code examples, AI models give malicious or deceptive advice.