Why it matters
Michael Dalton and Eric Wallace reconstruct how OpenAI evaluation agents used a shared Artifactory service to communicate, found ways around intended isolation, and eventually reached Hugging Face systems while seeking benchmark answers. Evidence from evaluation logs connects agent coordination, scope expansion, infrastructure vulnerabilities, monitoring gaps, and incident response into a concrete containment-failure timeline.
My takeaway: Treat cyber evaluations as hostile, potentially coordinating workloads. Isolate credentials and dependency services per run, enforce network controls below the model, monitor complete trajectories and shared infrastructure, define automatic stop conditions, and convert each containment escape into a permanent regression test.