Unit 42 warns of attackers turning AI red team tools on victims

Palo Alto Networks' threat intelligence team, Unit 42, published research on 11 November 2026 on the use of AI-powered red teaming tools by attackers, focusing on a tool it names as ARTEX.

Red teaming tools are software packages built to simulate attacks so that defenders can test their own systems. The newer generation of these tools uses AI models to automate parts of the work, such as finding a way in or chaining steps together. Unit 42's post is titled "ARTEX and AI-Powered Red Teaming Tools: What Defenders Need to Know", and the company said it sets out what Unit 42 "discovered about attacker use of AI-powered red team tools, and how to mitigate these attacks".

The framing matters for security teams because it places these tools in a familiar category. Offensive security software written for legitimate testing has long been picked up by criminal groups, because the tool does not know who is holding it. Unit 42's description indicates it has observed that pattern repeating with AI-driven tooling, rather than describing a single intrusion at a single organisation.

The published summary of the research does not name any victim organisations, does not attribute the activity to a named group, and does not describe a specific breach, timeline or volume of affected systems. It also does not state which AI models or providers the tools rely on. Readers who need those details will have to work from the full post, which Unit 42 said includes mitigation advice.

For defenders, the practical question raised by research of this kind is detection. Tools that automate reconnaissance and exploitation tend to generate activity that looks like ordinary testing, because that is what they were written to do. Distinguishing an authorised red team exercise from an unauthorised one usually depends on knowing what your own testing schedule looks like, not on the tool's signature alone. Unit 42 said its post covers mitigation, though the summary does not list the specific controls it recommends.

The Agentic Times has not independently verified Unit 42's findings. No other vendor or agency has, at the time of writing, published corroborating research on ARTEX that we have been able to review, and Palo Alto Networks is the sole source for the claims above. Organisations evaluating the risk should treat the vendor post as a starting point and check whether the behaviours it describes are visible in their own logs, particularly around automated scanning, credential use and lateral movement that does not map to a scheduled internal exercise.

Security teams that run their own AI-assisted red team tooling may also want to review who can access it and whether its activity is logged separately from production traffic, so that authorised testing can be ruled out quickly during an investigation.