Iranian OT hackers add AI to their attack workflow

Published

Iranian hackers who go after internet-facing industrial controllers have moved artificial intelligence out of personal chatbot accounts and into their own operating infrastructure, according to new research from Recorded Future's Insikt Group. The same report counts nearly 3,000 servers running ARTEX, an open-source AI agent that carries out break-ins largely on its own, double the figure of a month earlier.

The context is the attack on French supermarkets. On October 11 and 12, 2026, someone changed defrost settings on refrigeration controllers at more than 700 retail locations in France, causing fires, equipment damage and food losses estimated in the tens of millions of euros. On November 9, France attributed the attack to Shahid Kaveh, a unit of the cyber command of Iran's Islamic Revolutionary Guard Corps (our coverage).

A familiar group widens its target list

Insikt Group tracks activity that likely belongs to an element of Shahid Kaveh as GreenJuliett, also known as CyberAv3ngers and Bauxite. The researchers are careful to say they have not independently tied GreenJuliett infrastructure to the French refrigeration attacks. What they have seen, over the past three months, is the group continuing to target programmable logic controllers (PLCs), the small industrial computers similar to the exposed controllers the French stores used, through at least October 30.

The target set is also growing. Between August and October, Recorded Future observed GreenJuliett-linked infrastructure scanning or trying to exploit Fortinet FortiGate firewalls, the edge devices that often sit in front of operational technology (OT) networks, in Bahrain, Israel, Jordan, Kuwait, Saudi Arabia and Qatar. Its network data suggests the group is also working to add IoT and CCTV devices.

From a banned ChatGPT account to an in-house AI gateway

In October 2024, OpenAI banned accounts it linked to CyberAv3ngers, which had used ChatGPT to research default PLC passwords and draft scripts against industrial protocols such as Modbus TCP. The 2026 picture is different. In October, Insikt Group found a configuration for LiteLLM, an open-source gateway that puts several AI models behind one interface, alongside self-hosted chat applications built for Persian-language users, on infrastructure it assessed as likely used by GreenJuliett-linked operators. It also saw servers in the group's anonymization network talking to infrastructure that hosts Anthropic's Claude, while stressing that it has not confirmed specific misuse.

The researchers' point is that AI shortens the slow parts of an OT campaign: finding vulnerable device types, testing credentials and writing targeting scripts.

ARTEX doubles in a month

Where GreenJuliett uses AI as an assistant, frameworks like ARTEX make it the operator. ARTEX is one of eight offensive AI frameworks Recorded Future tracks. Transcripts recovered from one exposed server show the agent issuing roughly 7,000 shell commands, recording more than 1,000 facts and filing 54 findings. Recorded Future counted 1,506 servers running ARTEX on October 21 and 2,978 on November 19, across 58 countries, noting these are observed servers, not confirmed attacks. Only about one in six stayed up for the whole 30 days, and roughly 90 new ones appear on a typical day. The FBI warned about criminal use of the same tool last week (our report).

Take controllers off the internet, fence in FortiGates

Insikt Group recommends removing exposed controllers from the internet (CERT-FR named Copeland XWEB, Danfoss AK-SC and AK-SM, and Dixell XWEB units in the French attacks), restricting FortiGate management interfaces to known addresses and keeping firmware current, hunting for GreenJuliett indicators in logs going back to at least August, and blocking connections from known AI framework servers when no authorized testing is under way. Because those servers churn daily, it argues a hand-updated weekly blocklist is stale within days, and suggests letting automation take narrow, reversible blocking actions on its own. The attackers are handing their workflow to machines one step at a time, and the defenders' answer, on this reading, is to do the same, starting with the safest step.

This briefing is provided by IntelFusions for informational and defensive purposes only. It is based on sources assessed to be reliable at the time of writing, and analytic judgments carry the confidence levels indicated. Indicators of compromise are defanged; re-arm them only in controlled environments. IntelFusions is not affiliated with the organizations named and makes no warranty as to completeness or accuracy.

Earlier in Nation-State

All Nation-State briefings

Latest briefings

Read the full analysis on IntelFusions