In brief

  • Applications of LLMs are a focal point, demonstrating both defensive capabilities in bug bisection and offensive capabilities in hardware IP piracy.
  • A newly introduced lightweight defense mechanism for Windows targets Link Following attacks effectively with minimal overhead.

  • Link Following attacks in Windows exploit symbolic links to redirect file operations, a technique not fully addressed by existing mitigations due to poor compatibility or limited applicability.

  • LinkGuard introduces a lightweight state-aware runtime guard targeting Windows systems.

  • LinkGuard introduces an average of 1% overhead in microbenchmarks and 3.4% overhead in real-world application workloads, adding a negligible 5 ms latency on benign file operations.

  • Bocheng Xiang, Yuan Zhang, Hao Huang, Fengyu Liu, Youkun Shi. “LinkGuard: A Lightweight State-Aware Runtime Guard Against Link Following Attacks in Windows File System.” NDSS — https://www.ndss-symposium.org/ndss-paper/linkguard-a-lightweight-state-aware-runtime-guard-against-link-following-attacks-in-windows-file-system/

LLMBisect: Breaking Barriers in Bug Bisection with A Comparative Analysis Pipeline

  • Traditional bug bisection struggles when the bug-inducing commit does not modify the same functions as the patch or when heuristics fail to capture vulnerability logic.

  • The proposed LLM-based multi-stage pipeline uses patch information and compares multiple candidate commits to identify the bug-inducing commit.

  • This approach yields an accuracy improvement of 60% compared to a baseline LLM-based solution.

  • Zheng Zhang, Haonan Li, Xingyu Li, Hang Zhang, Zhiyun Qian. “LLMBisect: Breaking Barriers in Bug Bisection with A Comparative Analysis Pipeline.” NDSS — https://www.ndss-symposium.org/ndss-paper/llmbisect-breaking-barriers-in-bug-bisection-with-a-comparative-analysis-pipeline/

LLMPirate: LLMs for Black-box Hardware IP Piracy

  • The use of LLMs in hardware design introduces the risk of hardware IP piracy, where designs are rewritten to evade piracy detection tools.

  • LLMPirate is an automated technique leveraging LLMs to generate pirated circuit designs that successfully evade detection from state-of-the-art tools.

  • The generated pirated netlists incur an average gate count overhead of approximately 50%, showing evasion is feasible at a reasonable cost.

  • Vasudev Gohil, Matthew DeLorenzo, Veera Vishwa Achuta Sai Venkat Nallam, Joey See, Jeyavijayan Rajendran. “LLMPirate: LLMs for Black-box Hardware IP Piracy.” NDSS — https://www.ndss-symposium.org/ndss-paper/llmpirate-llms-for-black-box-hardware-ip-piracy/