Felony Bench

Felony Bench

A benchmark you really don't want models to be saturated with.<br>Learn more ↓

Score<br>↖ Most illegalLeast illegal ↘

Anthropic

OpenAI

Google

Meta

Moonshot

Scores indicate count of illegal activity. Higher is... you decide.

CompanyFeloniesDescriptionDateSource

Anthropic<br>Unauthorized use of GitHub credentials; Supply-chain attack on Dependabot; Social-engineering email campaign; Exposed a malicious DNS server to the public internet<br>8/4/2026<br>AISI ↗

OpenAI<br>Unauthorized use of GitHub credentials; Exposed a malicious DNS server to the public internet<br>8/4/2026<br>OpenAI/AISI ↗

OpenAI<br>Confirmation for one of four companies via Irregular's misconfigured CTF eval(?)<br>8/4/2026<br>OpenAI ↗

OpenAI<br>Compromised internal accounts at four companies<br>7/31/2026<br>Reuters ↗

Anthropic<br>Compromised internal accounts at three companies<br>7/30/2026<br>Anthropic ↗

OpenAI<br>Compromised Hugging Face<br>7/21/2026<br>OpenAI ↗