Illustration of a small robot moving through a maze of doors on its own, representing autonomous AI-driven cyber attacks
Use promo code “TCH15” to get 15% off on checkout.

Google expects fully autonomous cyber attacks to emerge within three to six months, after its Threat Intelligence Group observed a “paradigm shift” in how criminal and state-linked groups use agentic hacking frameworks over the past quarter.

Key Takeaways
  • Google’s Threat Intelligence Group expects fully autonomous, AI-driven cyber attacks within three to six months, after seeing criminal groups adopt agentic frameworks such as ARTEX, CAIRN and PentAGI.
  • Google found one framework that compared software before and after a patch, drafted a working exploit and fired it, with no human involved at any step.
  • Keeper Security’s Shane Barney says only 28 per cent of organisations can spot credential misuse within minutes, even as 38 per cent name AI-enabled malware as their top AI-related concern.

AI agents that scout, breach and adapt on their own

The shift Google is tracking centres on agentic frameworks with names like ARTEX, CAIRN and PentAGI, which assign AI sub-agents distinct jobs within a single attack: one scouts a target network, another breaches it, a third escalates privileges and moves laterally, and others write single-use attack code on the fly rather than reusing known malware. Google’s researchers say this lets an attack adapt mid-operation in ways that used to require a human operator watching and redirecting each step.

In one case the group traced, a framework compared a piece of software before and after a security patch was released, worked out what the patch had fixed, drafted a working exploit for the gap and fired it at a target, all without a person involved at any stage. Google’s researchers describe that kind of patch-to-exploit turnaround, done entirely by machine, as the clearest signal yet that fully autonomous attacks are close.

Still built on stolen credentials and unpatched systems

“The underlying drivers are textbook: stolen credentials and unpatched systems. What has changed is who does the work,” says Shane Barney, Chief Information Security Officer at Keeper Security.

Barney’s point is borne out in Keeper Security’s own 2026 research: only 28 per cent of organisations say they can detect credential misuse within minutes, and 38 per cent now name AI-enabled malware and automated exploitation as their top AI-related security concern. The weaknesses autonomous agents exploit, in other words, are the same ones security teams have struggled with for years, now being found and used at machine speed rather than human speed.

Closing the open door before an agent arrives

Barney argues the practical response is not a new category of defence but stricter discipline on the basics: privileged access management, just-in-time access that limits how long credentials stay valid, and automated rotation of secrets so a stolen key expires before it can be reused. The goal, he says, is “ensuring an unpatched server is a dead end, rather than an open door” — because an autonomous agent that finds no way through moves on, while one that finds a gap left for convenience will exploit it as readily as any human attacker would.

Author

Discover more from techcoffeehouse.com

Subscribe now to keep reading and get access to the full archive.

Continue reading