Palo Alto Launches Multi-Model AI Cyber Defence

Palo Alto Networks is turning frontier AI models into a continuous offensive-testing layer for enterprise security.

What happened

The company launched a subscription service that uses multiple AI models to search customer web applications, APIs, cloud infrastructure, source-code repositories and networks for vulnerabilities and potential attack paths.

Rather than relying on one model, the system combines findings across frontier and open-weight models.

In internal testing, Palo Alto says no single model found more than 40% of vulnerabilities, while findings from different models overlapped by less than 10%. Those figures are company-reported.

Why it matters

Security testing has traditionally depended on scheduled penetration tests and specialist teams. AI agents can potentially make that process continuous by probing systems more frequently and covering a broader surface.

The multi-model design is particularly interesting because it treats model diversity as a feature: different systems may identify different weaknesses.

That also reflects an uncomfortable reality for defenders. Attackers increasingly have access to the same capable models, so defensive security needs to operate at machine speed as well.

The bigger picture

Cybersecurity may become one of the first enterprise categories where autonomous agents are genuinely unavoidable. As offensive capability becomes easier to automate, defence will need equally automated discovery, validation and response. Palo Alto's launch shows established security vendors moving rapidly to turn AI agents into core infrastructure rather than optional assistants.