Research
BypassGym: A Benchmark for Evaluating AI Agents’ Endpoint Defense Evasion Capabilities
Publication date unknown · Discovered arxiv.org
Page published
Publication date unknown · First observed: 13 Nov 2026
Coverage timeline
Single-source research — one report is available.
Why it matters
BypassGym’s findings suggest that even lightweight AI agents can automate endpoint-defense evasion, giving defenders a concrete offensive capability to assess against their EDR controls.
BypassGym evaluates whether AI agents can generate and compile loaders that execute command-and-control shellcode without being blocked or terminated by EDR on real Windows virtual machines. Across five lightweight model backends, six EDR products, and three payloads, the authors report that DeepSeek-V4.1 Flash succeeded in 17 of 18 runs.