Wiz sets out what attacks run by AI agents look like

Cloud security vendor Wiz published a blog post on 29 October 2026 arguing that AI agents are changing the way attackers work, and setting out what it calls the "fingerprints" of agentic attacks. The post is a survey piece rather than a report of a single incident, so no named victim or breach is attached to it.

The core claim is simple. According to Wiz, AI agents are changing how attackers operate. The company frames this as a shift already under way rather than a forecast, and says the post covers real-world cases of agentic attacks, how to identify the traces they leave, and what they mean for the future of cybersecurity.

An agentic attack, in the loose sense the industry now uses, is one where software that can plan and act on its own carries out some of the work a human operator would otherwise do by hand. That is a change of tempo and scale rather than a change of technique, and it is the reason defenders have started asking whether their existing detection logic still fits. Wiz's post is pitched at that question: it offers identification guidance, which implies the company believes these attacks can be distinguished from conventional intrusions by how they behave.

For security teams the practical interest is in that identification step. Detection rules written for human operators assume human pacing, human error and human working hours. If an agent is doing reconnaissance, enumeration or lateral movement, the shape of the activity in logs may differ. Wiz says its post addresses what those differences look like, though the published summary does not specify them.

The article should be read for what it is. It comes from a vendor that sells cloud security products, and vendor research on emerging threat categories tends to serve two purposes at once: informing defenders, and defining a category that the vendor is positioned to address. That does not make the observations wrong. It does mean the underlying cases matter more than the framing, and readers evaluating the claim should look at the specific examples Wiz cites rather than the headline assertion that attacks are becoming agentic.

Nothing in the material available indicates a new vulnerability, a disclosed CVE or an active campaign against a named target. There is no patch to apply and no indicator of compromise to search for. What there is, according to Wiz, is a pattern worth learning to recognise before it becomes routine.

The Agentic Times has not independently verified the cases described in the post, and the summary published by Wiz does not name the organisations involved. Readers wanting the detail will need to go to the source. We will cover specific agentic intrusions as and when they are confirmed by victims, vendors or regulators.