# Benchmark Released for Uncensored and Offensive Security AI Models

A new benchmark evaluates AI models that have been stripped of safety refusals for offensive security and hacking tasks. It measures how well uncensored models handle penetration testing and exploit-related prompts. The work matters as a yardstick for the risks and capabilities of unrestricted AI in cybersecurity.

**Importance:** 2/5

## Sources

### Technology
- [Hacker News](https://github.com/JoasASantos/Offensive-Security-AI-Models) — Tue, 29 Sep 2026 06:16:04 +0000