Towards a Systematic Risk Assessment of Deep Neural Network Limitations in Autonomous Driving Perception

The document is a collection of recent security-relevant ML/AI papers that identify high-impact vulnerabilities and propose mitigations across multiple domains. Key findings include: systematic risk assessment of inherent DNN limitations for autonomous-driving perception (linking HARA and TARA frameworks); methods to align model sensitivity with predictive uncertainty to reduce adversarial/ambiguity failures; Security-Recall Divergence in long-context LLM agents where prohibition-style constraints decay while requirement constraints persist; widespread unintended information disclosure in arXi