OpenAI has unveiled GPT-Red, an internal automated red-teaming model built to discover and exploit prompt-injection vulnerabilities at scale, and then use those findings to harden production models. The company confirms that it directly incorporated GPT-Red into the training pipeline for GPT-5.6, resulting in what OpenAI calls its most robust model to date against prompt-injection attacks. […]
The post OpenAI GPT-Red Uses Automated Attacks to Strengthen GPT-5.6 Against Prompt Injection appeared first on Cyber Security News.