Fudan's MATE brings rule-aware security auditing to mobile agent trajectories
jiqizhixin · x · 2026-10-14
Researchers from Shanghai Institute for Intelligent Innovation and Fudan University present MATE, accepted at USENIX Security 2026 — the first rule-aware security auditing model for mobile/GUI agent execution trajectories. Static rule matching is cheap but misses long-horizon context semantics, while general LLM judges are flexible but costly, slow, and leak privacy by sending full trajectories externally. MATE takes natural-language safety rules as direct input and jointly models them with the instruction and trajectory, outputting risk category and audit result. Pipeline: collect workflows and security policies from 158 popular apps; knowledge-grounded trajectory synthesis with three-stage quality checking; data augmentation and SFT to build MATE and MATEBench; deployment via a trajectory adapter and policy retriever.
More from coding & agent
- Training repo-level code editing into weights to ditch heavyweight agent scaffolds — burkov · 2026-10-14
- UCLA's PANINI-DR hits 73.3% on BrowseComp-Plus, beating best deep-research harness by 16 points — M. Yigit Turali · 2026-10-14
- Mattermark creator launches StartupGraph: startup data API and MCP server for AI agents — Scobleizer · 2026-10-14
- 10 open-source repos to test your AI agent before your users do — victor_explore · 2026-10-14
- With superhuman coding agents, changing habits is still the hardest engineering problem — daniel_mac8 · 2026-10-14
- Robo Use launches: one API for agents across 32 robot embodiments — joemeno · 2026-10-14