OpenAI Built GPT-Red to Test Prompt Injection Attacks, GPT-5.6 Shows 6x Fewer Failures

According to OpenAI on July 15, the company built GPT-Red, an internal automated red-teaming model that tests AI systems for prompt injection vulnerabilities and helps adversarially train GPT-5.6 before deployment. OpenAI said GPT-5.6 Sol achieved 6x fewer failures on its hardest direct prompt-injection benchmark compared to its best production model from four months earlier. In a simulated test using a production AI vending machine scenario from Andon Labs, GPT-Red outperformed human red-teamers by changing prices and canceling another customer's order before OpenAI disclosed the flaws. Prompt injection occurs when malicious instructions embedded in prompts or external content alter a model's behavior unintentionally.
Disclaimer: The information on this page may come from third-party sources and is for reference only. It does not represent the views or opinions of Gate and does not constitute any financial, investment, or legal advice. Virtual asset trading involves high risk. Please do not rely solely on the information on this page when making decisions. For details, see the Disclaimer.
Comment
0/400
No comments