Skip to content
AI IntelligenceJul 16, 2026AI Intelligence
Article

OpenAI has built GPT-Red, an AI super-hacker that automatically tests its own models for security vulnerabilities.

GPT-Red finds successful attacks in 84% of test scenarios, while human red teams manage only 13%. This makes models like GPT-5.6 much more robust against cyberattacks.

AI-generated: summaries written by AI from the linked sources. How we use AI

AI-generatedSource: MIT Technology Review
01

Source Brief

OpenAI has built GPT-Red, an AI super-hacker that automatically tests its own models for security vulnerabilities. GPT-Red finds successful attacks in 84% of test scenarios, while human red teams manage only 13%. This makes models like GPT-5.6 much more robust against cyberattacks.

02