OpenAI’s GPT-Red Automates Prompt Injection Testing to Harden GPT-5.6 Sol

OpenAI has developed GPT-Red, an internal automated red-teaming model designed to identify and mitigate prompt injection vulnerabilities in its GPT series of models. GPT-Red simulates adversarial attacks by iterating through prompts to achieve malicious goals such as exfiltrating sensitive data or injecting malicious scripts, thereby improving the robustness of GPT-5.6 Sol. The model is used in adversarial training to harden production models against both direct and indirect prompt injections, significantly reducing failure rates compared to previous versions.

AI-extracted · verify before operational use