OpenAI has introduced GPT-Red, a specialised artificial intelligence model developed to improve cybersecurity testing by automating parts of the red teaming process. The announcement was made by the company’s security team on July 15, highlighting the model’s role in identifying vulnerabilities before they reach users.
Red teaming is a security practice in which experts simulate real-world cyberattacks to evaluate the resilience of software, systems and AI models. GPT-Red has been created to assist OpenAI’s internal security teams by accelerating this process and uncovering potential weaknesses more efficiently.
According to OpenAI, GPT-Red played an important role in the development of GPT-5.6 Sol, the company’s latest and most advanced ChatGPT model. The security-focused AI was used to help identify vulnerabilities and strengthen defences before the model’s release.
One of GPT-Red’s primary capabilities is detecting and testing against prompt injection attacks, a technique in which hidden or malicious instructions are embedded within prompts to manipulate an AI model into producing unintended or unauthorised responses.
OpenAI said the model was trained using reinforcement learning, enabling it to perform security testing more automatically. By simulating attack scenarios at scale, GPT-Red can help researchers discover and address security issues earlier in the development process.
The company emphasised that GPT-Red is not available to the public and is intended solely for internal security operations. Although the model can perform tasks that resemble hacking techniques, its purpose is defensive rather than offensive.
OpenAI explained that while human-led red teaming remains an essential part of AI safety testing, manual evaluations alone cannot keep pace with the growing complexity and scale of modern AI systems. Automating portions of the process allows security teams to generate and analyse significantly more attack scenarios than would be possible through manual testing alone.
The company also noted that as artificial intelligence becomes more capable, it can be used to create increasingly sophisticated cyber threats. Developing AI systems that can proactively identify and mitigate those risks has therefore become a key part of OpenAI’s security strategy.
The introduction of GPT-Red reflects a broader industry trend of using artificial intelligence not only to build advanced models but also to improve the security and reliability of those systems before deployment. As AI capabilities continue to evolve, automated security testing is expected to play an increasingly important role in protecting future models from emerging threats.
