Palo Alto Networks is a cybersecurity vendor whose relevance to AI policy runs along two lines: it uses frontier models in its own vulnerability-discovery process and publishes what that produces, and its threat-research arm, Unit 42, publishes on how attackers use AI. Both make it a recurring primary source on where AI sits in the offense–defense balance rather than an AI developer in its own right.
Use of frontier models in vulnerability discovery
Palo Alto Networks was a launch partner in Project Glasswing, Anthropic's defensive cybersecurity initiative announced in April 2026. On June 1, 2026 it disclosed that Claude Mythos drove the majority of findings in its "Patch Wednesday" advisory of May 13, 2026: 26 CVEs covering 75 issues across more than 130 products, against a typical month of fewer than five. The company said frontier models are "extraordinarily capable" at turning vulnerabilities into exploit paths, and estimated a three-to-five-month window for defenders to outpace attackers (Source: paloaltonetworks.com).
Anthropic's own Project Glasswing update recorded the same effect from the other side, reporting that the company's latest release contained more than five times its usual number of patches (Project Glasswing: An initial update). The episode is one of the clearest available measurements of what a frontier model changes inside an established vendor's disclosure pipeline, and it sits on the defensive side of the argument at Defensive AI Paradox.
Unit 42 threat research
Unit 42 is the company's threat-intelligence and incident-response arm. Its report of July 30, 2026, Chinese-Speaking Threat Actor Harnesses AI Models for Autonomous Cyberattacks, documents an opportunistic operator based in Zhuhai running DeepSeek through the open-source Hermes Agent framework as an autonomous offensive operator against exposed workflow-automation infrastructure (Chinese-Speaking Threat Actor Harnesses AI Models for Autonomous Cyberattacks). Its recovered session logs cut against the strongest reading of the campaign: the agent performed target selection and CVE research unaided, sampling roughly 100 of 25,209 exposed Chinese n8n instances and finding three on vulnerable versions, but compromised none, and across more than 460 attempted targets every confirmed impact traced to the actor's separate manual operations. Unit 42's characterisation of autonomous cycles as operationally viable with a narrow failure margin is a forward-looking judgment resting on target-side configuration having been what stopped the attempts. See Autonomous cyber-agents.
Unit 42 personnel are also cited on the economics of the shift: Sam Rubin's May 7, 2026 framing of AI's effect on attacker coding efficiency is among the positions collected at Policy Brief: Claude Mythos preview — three takeaways for AI policy.
Products
The company sells network security, endpoint detection and response, and security-operations products, and is listed among vendors shipping AI features in EDR and SIEM tooling (AI and Cybersecurity).
Relationships
- related: Anthropic — Project Glasswing launch partner and user of Claude Mythos in vulnerability discovery.
- supports: Autonomous cyber-agents — Unit 42's recovered sessions are among the empirical anchors on how much of an AI-enabled campaign is actually autonomous.
- related: Chinese-Speaking Threat Actor Harnesses AI Models for Autonomous Cyberattacks — the July 2026 report.
- related: AI and Cybersecurity, Defensive AI Paradox, Dual-Use Frontier AI.
- related: Claude Mythos Preview — the model behind its May 2026 advisory.