# OpenAI Enhances Model Security with GPT-Red GPT-Red helps protect AI models against cyberattacks. - **Source:** [MIT Technology Review](https://www.technologyreview.com) - **Published:** 2026-07-15 - **Topic:** [Data & AI](https://brifing.si/c/data-ai) - **Original article:** - **On Brifing:** https://brifing.si/s/9a0ea8ac-54ef-47c4-953b-778aaad16a20 ## What is GPT-Red? OpenAI has developed an AI model called GPT-Red. This model is used as a training partner to strengthen other models' defenses against cyberattacks. GPT-Red automates the red team simulations typically conducted by human test teams. ## Security of GPT-5.6 OpenAI announced that GPT-5.6, trained with GPT-Red, is the most secure version to date. GPT-Red identifies vulnerabilities in software systems, allowing these weaknesses to be patched. This prevents potential attacks when models are used in various tasks. ## Prompt Injection Attacks GPT-Red focused particularly on prompt injection attacks. In these attacks, a hacker can make an LLM perform unwanted actions, such as copying confidential information or generating harmful outputs. GPT-Red works on detecting and preventing these attacks across various scenarios. ## Self-Play Loop GPT-Red was trained in a self-play loop with other models. In this loop, GPT-Red attacked other models while they attempted to defend themselves. This process enhanced GPT-Red's attack capabilities and the defense mechanisms of the other models. ## New Attack Types GPT-Red discovered a new type of attack previously unseen by researchers: fake chain of thought. In this attack, GPT-Red inserts a fake entry into another model's chain of thought, tricking the model into acting on false information. Such attacks pose a serious threat to model reliability. ## Future Security Testing OpenAI designed GPT-Red for future security testing. As more capable models emerge, GPT-Red will be able to discover new types of attacks. This is seen as a critical step in continuously enhancing the security of AI models. ## Sources covering this story - https://www.technologyreview.com/2026/07/15/1140514/meet-gpt-red-an-llm-super-hacker-openai-built-to-make-its-models-safer/ — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._