In May 2026 the Tejas Cyber Network, a vendor-neutral community of security leaders, published a curated canon of 61 AI security and safety benchmarks: the ones frontier labs and the US and UK AI Safety Institutes actually cite. This directory lists all 61, one short page per category, with where Paladin stands on each.
Most of the canon measures the model: what it knows, what it can do, what it refuses. An enforcement layer in front of the model can only be scored on a narrow slice, the agent-security benchmarks, and that is the slice our results live in. We publish the whole list anyway. A buyer comparing vendors should be able to see which recognized measurements apply to this product category at all, and which of those we have run.
Entries marked top 12 are the twelve the canon finds most consistently cited across 2025 and 2026 frontier system cards.
The canon is maintained by Anshu Gupta and published by the Tejas Cyber Network. Categories, membership, and the top-12 ranking are the canon's. The one-line descriptions and the Paladin column are ours. We fetch-verified every link before publishing and corrected two that pointed at the wrong papers. The canon. Its GitHub repository.
Published Paladin results and their caveats: Paladin benchmarks.