OpenAI releases a model built to find and exploit software flaws

What did OpenAI release?

OpenAI rolled out GPT-6 Cyber on Thursday. It is an AI model designed to find software security holes and build working attacks on them. For now, only a limited group of vetted customers can use it.

Why it matters: OpenAI rated the model's cybersecurity capabilities as "Critical" under its Preparedness Framework. The company said the model beat its previous cyber model on ExploitGym, a test of whether a model can exploit a specified flaw. OpenAI also said it did better at staying within authorized limits.

What flaws did the model find?

OpenAI said its researchers used the model to find 81 previously unknown flaws in Firefox, Mozilla's web browser. Mozilla rated them as high severity and has begun releasing fixes, OpenAI said.

In the Linux kernel, the core of the Linux operating system, OpenAI said the model helped build working exploits for flaws that included 588 previously unknown ones. The company said it sent its findings and proposed patches to the people who maintain the code.

How did Cisco and Palo Alto Networks shares trade?

OpenAI said in a post on X that security teams at Mozilla, Palo Alto Networks (Nasdaq: PANW) and Cisco Systems (Nasdaq: CSCO) already use the model.

On Friday, Cisco shares slid 2.0% to $93.85, and the company lost $7.4 billion in market value. Palo Alto Networks fell 0.3% to $313.84. OpenAI's statements did not address either stock.

Both stocks sit well below their 52-week highs. Cisco is 27.8% under its peak, and Palo Alto Networks is 25.3% under its own.

What comes next?

What’s next: OpenAI said it plans to widen access gradually to all customers of Daybreak Red, its program that gives vetted organizations advanced AI security tools.

The open question is whether models like this become a product security companies sell or a rival to them. Cisco is scheduled to report quarterly results on Nov. 12, which may offer the company's first comments on that.