RACER is presented, a model-level repair framework motivated by a key observation: backdoors induce abnormal layer-to-layer evolution in internal representations, which is modality-dependent, concentrating primarily in the token region encoding the trigger features that the backdoor model actually relies on.
Jia-Li Wei, Ming Fan, Ming-Kun Zhang et al.· arXiv.org· 0 citations
Multi-turn attacks on agentic systems can compose individually permissible actions into harmful outcomes, challenging defenses that assess actions or states in isolation. We show that such attacks leave a detectable signature in the agent's internal representations: harmful behavior emerges as an accumulated representa...
Hao-Yu Wang, Wei Zhao, Ye-Di Zhang et al.· 1 citation
PolicyMem is introduced, a geometric policy memory that externalizes natural-language policies as reusable geometric memory objects represented by low-rank subspaces in a shared representation space that achieves state-of-the-art unsafe behavior detection while enabling effective policy attribution, rewriting, and post...
Yuan-Chen Bei, Zheng-Zhang Chen, Yan-Jun Zhao et al.· 0 citations
An automated transformation pipeline that converts existing safety designs into reusable safety skills and establish a community-driven safety skill library is developed and results demonstrate the potential of stage-specific safety skills as a scalable and composable foundation for building resilient and trustworthy a...
Averaged over five seeds, MemGuard achieves the best success metric and lowest average steps in all 16 backbone-benchmark settings, improving over ReasoningBank, the strongest prior baseline among the memory methods the authors evaluate.
Hao-Yu Wang, Guang-Yuan Dong, He Liang et al.· 4 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.