OpenAI Launches LLM Super Hacker GPT-Red to Boost Model Security

OpenAI built GPT-Red, an LLM super hacker that acts as a sparring partner to help its other models improve defenses against cyber attacks. Last week, OpenAI released GPT-5.6, the latest version of its flagship LLM. OpenAI says GPT-5.6 became the most robust version yet after training with GPT-Red. GPT-Red automates security testing, enabling large-scale, continuous adversarial training. For developers using GPT-5.6, this means the model is inherently more resistant to common attack vectors like prompt injection, reducing the need for extra application-level safeguards. Security teams can also adopt automated approaches like GPT-Red to continuously improve model threat resistance.