Ransomware affiliates ran intrusions with AI agent framework

Security researchers have uncovered an intrusion ecosystem built around artificial intelligence agents and run by affiliates of the INC and The Gentlemen ransomware groups, according to a report published on 23 October 2026 by The Hacker News. At the centre of the operation sits what the researchers describe as an "Autonomous Red Team Kill Chain Framework" referred to as Phantom.

The discovery came from exposed infrastructure belonging to the operation, the report said. That is the recurring pattern in this kind of research: the attackers leave a server or control panel reachable from the internet, and investigators work backwards from it to reconstruct how the tooling was used.

According to the report, the threat actor behind the framework is assessed to be Russian-speaking and operates both as an intrusion operator and as a data-extortion operator. In other words, the same group gains access to victim networks and then runs the extortion that follows, rather than simply selling access on.

The name Phantom gives is own summary of the design intent. A kill chain is the sequence of steps an attacker moves through, from first access to final objective. A framework that claims to automate that chain is one that tries to hand individual steps to software agents, rather than having a human operator type each command. The researchers' use of the term "autonomous red team" points to tooling modelled on the offensive security kits that legitimate testers use, repurposed for criminal intrusions.

Beyond those details, the published material is limited. The report does not name the underlying AI models, the specific tasks delegated to agents, the number of victims, or the period over which the framework was used. No vendor has been named as the source of the technology, and no fix or mitigation has been announced, because the issue here is misuse of attacker-controlled tooling rather than a flaw in a particular product.

For defenders, the significant claim is not that criminals are experimenting with AI, but that a ransomware affiliate ecosystem has built a structured framework around agent-driven activity and used it in real intrusions. That suggests detection teams should expect intrusion activity that moves faster and more consistently than hands-on-keyboard operators, though the report stops short of quantifying any change in speed or success rate.

The Agentic Times will report further details as researchers publish them. For now the incident is best read as an early, partially disclosed look at how offensive automation is being packaged and shared inside ransomware affiliate networks.