An unsupervised stage-level diagnostic is developed that localizes silent failures to the responsible operation at 76.2% accuracy and establishes circuit-level signatures as a practical handle for monitoring agent memory and guiding its structural design.
Xutao Mao, Jin-Man Zhao, Gerald Penn et al.· arXiv.org· 1 citation
Results show that preserving already-correct work under unsupported accusation is a distinct safety challenge for long-lived agents, and introduces CAVE-Bench, a benchmark of 365 agentic tasks across six domains built around opaque tasks.
Xu-Tao Mao, Rui Qian, Long-Xiang Wang et al.· 0 citations
This work proposes Targeted Anti-obfuscation with Mechanistic Enforcement (TAME), which uses Sparse Autoencoders (SAEs) to combine behavioral feedback with targeted suppression of template-associated activations during RL, providing a path from behavioral monitoring to representation-level oversight for more auditable...
Xu-Tao Mao, Jianing Zhu, Jin-Man Zhao et al.· 0 citations
This work proposes SafeEvolve, which checks each skill when it is written and when a later task retrieves it, and introduces SkillMisevo-Gym, which clears every task session but keeps the skill library, so each of these steps can be checked separately.
This work presents AHA, a falsifiable discovery loop that proposes a vulnerability hypothesis, constructs a falsifier, instantiates a valid attack, executes it in a sandboxed harness, reflects on the trajectory, and promotes confirmed findings into a Vulnerability Concept Graph (VCG), which provides an auditable artifa...
The Personal Agent Sycophancy Benchmark (PASB) is introduced, a 1,600-task benchmark that traces whether a conversational claim is accepted, written into durable agent state, and reused in a later neutral query, and shows that agent sycophancy is fundamentally a state-writing governance problem.
Xutao Mao, Liang Zhao, Leyao Wang et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
