OpenAI Launches GPT-6 Cyber to Detect Security Flaws and Help Researchers Fix Them

OpenAI has introduced GPT-6 Cyber, an artificial intelligence model designed to identify cybersecurity vulnerabilities, assess how they could be exploited and help researchers fix them before malicious actors can take advantage of them. The company initially made the model available on Thursday to a limited group of vetted customers, classifying it as “Critical” for cybersecurity under its safety framework. OpenAI said its researchers used the model to identify 81 previously unknown vulnerabilities in Mozilla Firefox, which Mozilla classified as high severity and has begun addressing through security updates. The model also helped develop functional exploits for 948 Linux kernel vulnerabilities, including 588 that had not been identified earlier.

OpenAI has also disclosed 304 critical Linux security findings to maintainers, along with proposed patches. The company upgraded GPT-6 Cyber's cybersecurity risk classification to “Critical” from the “High” rating assigned to its predecessor, GPT-5.6 Cyber. OpenAI said it has observed increasing attempts to misuse artificial intelligence, including by malicious cyber actors and groups attempting to distil its models. GPT-6 Cyber is currently available to select customers through Daybreak Red, the company's access programme for advanced authorised security activities such as penetration testing and exploit development. Access will be extended to all approved Daybreak Red customers over the coming weeks, while GPT-5.6 Cyber is being made available to all customers of the programme.