OpenAI Uses AI Red Team to Strengthen GPT-5.6 Against Prompt Injection Attacks

OpenAI's new automated red-teaming model, GPT-Red, uncovered vulnerabilities used to make GPT-5.6 more resistant to prompt injection attacks. The AI company announced the security improvement on July 16, 2026, strengthening its latest model against adversarial inputs that could manipulate its behavior or extract sensitive information.