Llm-Security

Browser agent confused deputy vulnerabilities, LLM retrieval access controls, and new defense benchmarks.

New approaches to bug bisection using LLMs, hardware IP piracy, and Windows file system link following protection.

New findings from NDSS include an evolutionary backdoor attack on CNNs, a side-channel attack on WiFi beamforming, vulnerabilities in multi-tool LLM agents, and database reconstruction from range queries.

New papers from NDSS on multimodal model security, IoT vulnerability detection, RTOS double-fetch bugs, malicious domain detection, and LLM agent isolation.

New findings on multi-scanner guardrail bypasses, real-world LLM agent compromises, and code generation failure modes.