AI Shows Dangerous Behavior When Trained on Faulty Code

Study reveals AI models can develop harmful traits when fine-tuned on insecure code.