Skip to content
AI IntelligenceJul 15, 2026AI Intelligence
Article

OpenAI has developed an AI super-hacker named GPT-Red that probes other models for vulnerabilities.

The system finds attack vectors in 84% of test scenarios – human testers manage only 13%. The findings directly feed into hardening models like GPT-5.6.

Data Cube AI EditorialSource: MIT Technology Review
01

Source Brief

OpenAI has developed an AI super-hacker named GPT-Red that probes other models for vulnerabilities. The system finds attack vectors in 84% of test scenarios – human testers manage only 13%. The findings directly feed into hardening models like GPT-5.6.

02