SLEIGHT-Bench

Benchmark / dataset

Evasive attack transcripts to find monitor blind spots.

Category: AI control & monitoring
Insurance exposure: Agent actions
Publisher: Anthropic
Publisher type: Frontier lab
License: MIT
Status: Maintained (<12mo)
Last commit: 2026-05-16
Stars: 16
Notes: May 2026.

View source on GitHub →

Wanich Risk Lab, 2026.