RELEASES

Release Notes

Track the evolution of the Agent Safety Benchmark and evaluation infrastructure.

MAJOR2026-02-01

First public release of the Agent Safety Benchmark. 100-point evaluation framework covering eight dimensions of agent behavior.

BETA2026-01-15

Public beta of the evaluation pipeline. Leaderboard, verified results, and API access for early partners.

FEATURE2026-01-01

Adversarial stress-testing pipeline integrated into the core evaluation suite. Exploit → Fix workflow enabled.

PREVIEW2025-12-15

New task suites for chain-of-thought reasoning under constraints and multi-step tool orchestration.