179 blocked tool calls were mostly false positives; a re-verification loop is coming
yb2698 · x · 2026-11-09
safeagents blocked 179 tool calls during training (0.04% of total), but on review many were false positives — e.g. the model harmlessly trying a random script on a benign site got caught.
Two mitigations are planned: loosen the safeagents plugin's conditions for such hosts, and feed the 179 blocked calls to several top models to pick a decision model that re-verifies blocked calls in the loop, minimizing false positives. A v2 of the benchmark is also on the way.
More from coding & agent
- WonderDuck: open-source robot duck ships full sim-to-real pipeline — philfung · 2026-11-09
- Agent Security Incidents Need Tamper-Proof Audit Trails, Khala Research Report Argues — 0xSammy · 2026-11-09
- SenseTime open-senses Code2Video: an open model that makes editable videos via code — HeyZoyaKhan · 2026-11-09
- Alexandr Wang: nobody writes code anymore — orchestrating agents is the job — 2C_ornot2C · 2026-11-09
- How a cloud AI agent joined a home network via Tailscale and drove a radio — Dave_Maynor · 2026-11-09
- Agent wiped production DB in 9 seconds — dev says keep humans in CD loop — Pavan_Belagatti · 2026-11-09