OpenAI introduced GPT-6 Cyber, a model designed to detect software vulnerabilities and aid developers with remediation guidance.

Model capabilities and results

Researchers using the model identified 81 previously unknown high-severity issues in Firefox, many of which Mozilla has already patched.

On Linux, the model assisted in producing 948 working exploits, demonstrating how discovered flaws can be translated into practical attack code.

How findings were handled

According to the report, Mozilla received details on the identified issues and implemented fixes for numerous bugs, following standard vulnerability disclosure practices.

The creation of working exploits underscores the need for coordinated patching and risk assessment when automated tools surface exploitable defects.

Access and restrictions

Access to GPT-6 Cyber is limited to participants approved by OpenAI, with the company controlling who can use the model for research or testing.

OpenAI states the pathway for access is gated to manage potential misuse while enabling security research within an approved community.

Implications for security research

The deployment illustrates how advanced generative models can accelerate vulnerability discovery and exploit development, changing workflows for both researchers and vendors.

Security teams and maintainers may need to adapt disclosure and mitigation processes to address findings that originate from automated analysis at scale.