Research
AgenticBench: leaderboard
Publication date unknown · Discovered agenticbench.org
Page published
Publication date unknown · First observed: 27 Oct 2026
Coverage timeline
Single-source research — one report is available.
Why it matters
AgenticBench helps defenders compare coding-agent security controls, although single-run measurements and explicitly unmeasurable outcomes limit the assurance its scores provide.
AgenticBench publishes a rolling security leaderboard evaluating AI coding agents' network egress, telemetry, workspace boundaries, and handling of secrets—not the security of generated code. The page reports nine tests across 25 harnesses, using single runs in fresh containers and evidence captured from system calls, model requests, and network traffic.