Palo Alto Networks uses AI models to hunt security holes

Palo Alto Networks, one of the world's largest cybersecurity companies, launched a new service on September 22, 2026, called Unit 42 Continuous Frontier AI Defense. The service uses advanced AI language models to continuously test companies' IT systems for security holes.

Behind the service are, among others, Anthropic's Claude Mythos 5 and OpenAI's GPT-5.6-Cyber, models of the same type as the chatbots Claude and ChatGPT, but specially adapted to find vulnerabilities in software and networks. The models are combined with Palo Alto Networks' own open-source models in what the company calls a multi-model architecture.

Specifically, the service continuously tests companies' websites, applications, APIs (the digital interfaces programs use to talk to each other) and cloud systems. When the AI finds a weakness, it also tries to confirm whether it can actually be exploited by an attacker, and it points to concrete solutions that the company's IT staff can use to fix the flaw.

The background is that criminal hackers are increasingly using AI themselves to find and exploit security holes. According to Palo Alto Networks, AI can shorten the time from when a hole is found to when it is exploited from weeks down to a few hours, in some cases up to 97 percent faster. The company's security division Unit 42 therefore wants to give companies a defense that can keep up at the same pace.

Sam Rubin of Unit 42 says AI has given attackers an advantage because companies typically defend themselves at human speed, while attacks now happen at machine speed. A representative from Anthropic says that during testing, their model found more than 10,000 serious vulnerabilities, including flaws that had survived decades of human review.

The service is sold globally as an annual subscription, with the price depending on which combination of AI models the customer chooses to use.

What this means for you

If you are responsible for IT or work at a company that uses Palo Alto Networks, the service can help find security holes faster than attackers can exploit them. You can ask your IT department whether your cloud systems and APIs are already covered by this kind of continuous AI check. If you are a developer or in charge of operations yourself, it may be an argument for prioritizing faster patching of vulnerabilities when they are found.