False Security Confidence in Benign LLM Code Generation

Xiaolei Ren

Prior work has demonstrated that functionally correct yet vulnerable outputs arise systematically in threat-oriented settings, where adversarial or implicit channels are used to in…