Felony Bench
AssessTools
A benchmark that measures AI agent incidents by harmful or illegal real-world actions.
Why it's here
Placed in Assess: 1 article(s) of evidence from 1 source(s), led by security coverage, with 0 in the last 30 days. Confidence 24%. Low accumulated evidence, so it defaults conservatively pending more signal.
Evidence (1)
- 7Hacker News·8/21/2026securityFelony Bench ranks AI agent security incidents
Felony Bench is a benchmark that scores AI models and agents by the number of illegal or harmful real-world incidents they have been associated with, rather than by task performance. The page lists recent cases involving Anthropic, OpenAI, Meta, Google, and Moonshot, and explains that only unique incidents affecting third parties are counted.