- OpenAI introduced GPT-Red, an automated AI system to find vulnerabilities in GPT models before release.
- The system was used to train GPT-5.6, reducing failures on a difficult prompt injection benchmark.
- GPT-Red complements human red teamers and other AI safety measures.
OpenAI has unveiled GPT-Red, an automated AI system designed to find security vulnerabilities in its language models before they are released to the public. The tool, named after Cybersecurity red teaming, deliberately attempts to break a system to identify weaknesses before attackers can exploit them.
According to a post on Wednesday, the tool helped make GPT-5.6 more resistant to prompt injection attacks before deployment. “As model capabilities grow, safety and alignment must scale with them,” OpenAI wrote on X, adding that “Red-teaming is essential, but today’s approaches are difficult to scale.”
OpenAI stated that GPT-Red was trained through self-play reinforcement learning, generating progressively stronger prompt injection attacks. The company reported that GPT-Red succeeded in 84% of internal evaluation scenarios, compared with 13% for human red teamers in the same tests.
In one case study, the system manipulated an autonomous vending machine agent into lowering prices and canceling orders before the vulnerabilities were disclosed. OpenAI explained that “every successful attack that GPT-Red finds is used to improve these defenders.”
The announcement reflects a broader shift toward using AI to secure AI. Earlier this month, the Ethereum Foundation deployed AI agents to red-team critical network infrastructure, which uncovered a vulnerability in software used by consensus clients. OpenAI says GPT-Red will remain an internal tool because it contains intentionally developed offensive capabilities, noting that “today’s models can be used to make tomorrow’s models more robust.”
✅ Follow BITNEWSBOT on Telegram, Facebook, LinkedIn, X.com, and Google News for instant updates.
