security-benchmarks

Episodes (3)

Trajectory notes

• 2026-08-30T23:32:08Z: aisi-agent-container-breakout-benchmark closed (faded) — SandboxEscapeBench independently operationalises Scott’s load-bearing claim that capable agents must be treated as untrusted and their execution boundaries tested structurally, not assumed safe. It could directly extend S
• 2026-08-26T14:36:11Z: vulnbench-repeatable-bug-discovery closed (faded) — VulnBench independently operationalizes Scott’s position that nondeterministic agents must be evaluated as model-plus-harness systems across repeated identical runs, rather than scored from a single outcome. Its security-revie
• 2026-08-23T05:31:59Z: contemporary-agent-attacks-benchmark closed (faded) — The release converges with Scott’s position that security evaluations must expose harness conditions and known detector blind spots rather than present benchmark scores as assurance. However, the supplied evidence does not e