When affiliates of INC and The Gentlemen ransomware groups ran a live intrusion campaign through commercial AI models, it took a third-party security vendor — not the labs behind those models — to expose it. That gap raises an important question: Should AI labs be catching criminal use of their own models, or are they relying on outside researchers to do it instead?
Large language models have changed ransomware economics, giving groups the capacity to scale and automate in ways once limited by time, tooling and budget.
Speaking to TechTarget, Dr. Hugh Thompson, program chairman of the RSAC Conference, described ransomware as a business with a sales funnel, where AI is “collapsing the cost of the top of that funnel.”
The objectives behind these attacks — extortion, financial gain and reputational harm — haven’t changed. What has changed is how easily AI lets attackers reach them. The question is how much responsibility the labs behind these models should bear for the misuse of their creations.
Weaponizing AI in ransomware campaigns
The FBI’s Internet Crime Report 2025 recorded 63 new ransomware variants last year, with Akira, Qilin, INC Ransom/Lynx/Sinobi, BianLian and LockBit among the most frequently reported. Acronis researchers estimated that 80% of ransomware-as-a-service operators actively promoted AI and automation features to affiliates and customers — and INC is a case in point.
According to a recent threat advisory published by SOCRadar, affiliates of INC and The Gentlemen RaaS operation are running what researchers call an “AI-driven intrusion ecosystem.”
SOCRadar’s threat research unit found a publicly exposed command-and-control server showing intrusion campaigns being run through OpenCode, an open source AI coding agent, using commercial models — including Anthropic’s Claude and Moonshot AI’s Kimi K3 — accessed via OpenRouter.
The ransomware operators also built their own tool, Phantom, an “autonomous red team kill chain framework,” to track victims and steal credentials, logging more than 8,000 AI messages across 133 sessions and targeting at least 62 named organizations.
One attributed OpenRouter account alone accounted for more than $23,000 in AI API spend — a sign that AI credits, not traditional tooling costs, are becoming a line item in ransomware operations.
SOCRadar assessed the ecosystem as “AI-executed, human-approved” with high confidence: the human operator directs and approves, while the AI stack plans reconnaissance, selects techniques, generates commands, triages stolen data and drafts reports.
That shift in cost structure is reshaping the ransomware market more broadly. Flashpoint estimates the cost of purchasing initial access to a victim’s system has fallen by 69% on the ransomware black market.
“AI gives attackers scale, personalized lures for initial access and the ability to spin up infrastructure to execute attacks,” Thompson told TechTarget. “The result is that a single operator can go after multiple targets at once and is also likely to have a higher success rate with those targets.”
Steve Winterfeld, advisory CISO at Akamai, agreed, saying AI has resulted in “hyper-targeted phishing attacks on steroids,” adding that AI-driven automation is “industrializing the path to compromise and making it easier to scale criminal operations.”
Should frontier AI labs take responsibility?
When evidence shows that AI models developed by frontier labs, open source developers and smaller projects are being abused by ransomware operators, the question of responsibility surfaces.
The INC and The Gentleman ecosystem was exposed by a cybersecurity firm, not an AI lab. It’s not the first time, either: In July, the Cybernews research team discovered an exposed server belonging to a ransomware affiliate of The Gentlemen that contained 3.1 TB of stolen data and an agentic AI harness running automated attacks.
Had the servers in both cases remained secure, this abuse of AI tooling might never have been discovered.
There’s also the question of obfuscation. Based on SOCRadar’s findings, the cybercriminals used OpenRouter’s API key gateway rather than going directly to the AI labs, meaning any traffic reaching each model would not have appeared as ransomware activity. When the model refused to comply with requests, the attackers reframed their prompts under “Authorized Red Team Assessments,” which appeared to result in AI compliance.
AI isn’t the first technology developed for ethical purposes to be seized and twisted into a malicious tool. Penetration testing software, built for defensive and ethical security testing, has long turned up in cybercriminals’ playbooks — and the idea of placing blame on these developers has always been an uncomfortable one. So, should frontier AI labs be any different?
AI labs, including Anthropic and OpenAI, recognize the potential for abuse in their agentic AI creations, and both organizations publish transparency reports on how they detect, counter and handle misuse. These labs, and others, maintain self-monitoring practices and continually refine their guardrails. OpenRouter, for example, says that unauthorized “red teaming” could result in account termination.
Anthropic has published reports on disrupting campaigns that exploit its models and on working with authorities, as highlighted in findings from an AI-orchestrated cyberespionage campaign in November 2025. OpenAI also publishes work on disrupting campaigns, though it tends to keep details of how it works with law enforcement vague.
There’s also little legal pressure or government opinion on how responsible frontier AI labs should be when their models are misused. California’s SB 53 legislation requires frontier AI labs to report potential critical safety incidents, but it’s unlikely that a ransomware operator quietly accessing Claude via an API gateway would meet that reporting threshold.
The question of responsibility remains unsolved. Fortra can’t eliminate every cracked copy of Cobalt Strike in the wild, but it can seek court orders to take infrastructure offline — and frontier AI labs have a comparable toolkit available to them: account termination, transparency reporting, tightened guardrails and cooperation with law enforcement.
Whether responsibility is ultimately laid at the feet of AI developers or not, it doesn’t change the risk enterprises face in the interim. AI is lowering the barrier to entry for ransomware and accelerating the speed at which these attacks unfold. According to Winterfeld, minimizing the impact when — not if — an attack occurs is one of the strongest measures companies can take.
“The best defense needs to be built around minimizing impact through AI-powered microsegmentation to stop lateral movement once adversaries get in,” Winterfeld said. “It stops the infection before it has a material impact. It’s how you stop a perimeter breach from turning into a full-scale ransomware crisis.”
Charlie Osborne is an investigative cybersecurity journalist with more than a decade of experience covering threats, data breaches, vulnerabilities and the wider security industry.