Jev-Guard: Policy tree routing to audit agent actions
Start here
An arXiv paper released on October 13, 2026 proposes Jev-Guard, a guardrail scheme for agent actions: it unifies safety classification and policy routing, selects the applicable policy on an editable policy tree based on the caller's audit context, and then performs safety classification accordingly. On 7,033 evaluation cases, it achieves 93.72% safety accuracy, 58ms average latency, and 93.10% hierarchical routing accuracy. After integration into the Hermes harness, its permission gate reduces the average review time per task from 4.80 seconds for the SFT-LLM judge to 1.33 seconds.
Generated by AI from reports · Updated 29 days ago
Attention and developments
Attention, hourly
Now 23, 0% in 24 hours·Peak 23 (Nov 6 03:00)
Related research
This event's paper
- Jev-Guard: Fast Routing and Auditing Agent Actions over Editable Policy Trees
arXiv · 2026-10-13 · Deep reading
Progress timeline
- arXiv: Jailbreaks, Prompt Injection, Poisoning & DefensesJev-Guard: Fast Routing and Auditing of Agent Actions on an Editable Policy Tree
Jev-Guard unifies safety classification and policy routing: it audits the context by caller on an editable policy tree to select the applicable policy, then performs safety classification accordingly. On 7,033 evaluation cases, it achieves 93.72% safety accuracy, 58ms average latency, and 93.10% hierarchical routing accuracy. Integrated into the Hermes harness, its permission gate reduces the average audit time per task from 4.80s with the SFT-LLM judge to 1.33s.