11/19 2026
331
Leak sites claimed nearly 3,000 ransomware attacks this summer, 20% to 40% more than in spring. Fully automated AI ransomware has yet to appear, but the barriers to phishing, negotiation, and going independent have fallen; one novice made over $800,000 in two weeks.
On November 14, Wired's cybersecurity column Kernel Panic reviewed a counterintuitive situation: ransomware attacks fully driven by AI have not yet appeared, yet both the number of attacks and the number of participants have risen at the same time. The judgment of multiple researchers is that AI has not yet replaced criminal groups, but has lowered the barrier to entry, allowing individuals who previously could only attach themselves to mature organizations to launch their own operations.
The numbers need to be taken with a grain of salt. Victims do not necessarily disclose incidents, and attackers exaggerate their results, so what researchers count is claimed attacks on leak sites, not verified cases. Even so, the tracking direction of multiple institutions is consistent: this summer such claims approached 3,000, estimated by some to be 20% to 40% higher than in spring. Cynthia Kaiser, former deputy director of the FBI's Cyber Division and now leading research at the anti-ransomware platform Halcyon, said the recent increase mainly comes from attacks outside the United States; the U.S. side is relatively stable but the absolute number remains high, and she believes part of it is driven by AI.
What AI takes over are the most trivial steps in the attack chain
Recorded Future threat intelligence analyst Allan Liska said the places where AI is used more in the ransomware attack chain are writing more convincing phishing emails and the negotiation stage, while successful groups are still largely operated by people. He added a caveat: these groups are testing AI extensively, and the situation could change quickly. Kaiser mentioned a specific scenario: scams in Japan have increased noticeably in recent months because foreign attackers can for the first time reliably generate decent Japanese content with AI. The language barrier disappears, and targets that were previously out of reach become reachable.
The tactics of mature groups are also being dismantled. Kaiser has seen more actors use AI to lure targets, find vulnerabilities, check their own code, and even compromise other groups' remote access tools to prepare redundant channels for the same victim. Traditionally, mature organizations carry out double extortion—encrypting systems and stealing data, then threatening disclosure; lone operators tend to do only data extortion, simplifying the process. Liska also provided a comparison: the number of attacks Recorded Future has recorded this year has already far exceeded all of 2025, but the number of attacks involving encryption may instead have declined year over year.
111 groups and a teenager who made $800,000 in two weeks
Image source: Wired
Ecosystem fragmentation explains the problem better than total growth. Gentlemen, Qilin, and Akira have been relatively stable groups in recent months, but a batch of small groups has emerged around them. Halcyon's count is that about 111 ransomware groups claimed to have launched attacks from July to September this year. In ReliaQuest's analysis, a new group calling itself Global Secret Group listed 48 victims, more than half posted on the same day; another called Storm listed 71 in 20 days. These claims do not necessarily all correspond to real intrusions, but they are enough to keep enterprise security teams busy verifying them.
ReliaQuest threat research director Brandon Tirado attributes the growth to the acceleration AI gives attackers, allowing them to no longer depend on ransomware-as-a-service providers and to launch their own operations. His team recently identified the operator behind a new group as an inexperienced teenager who made more than $800,000 in less than two weeks; the name was not disclosed. Kaiser is cautious about this batch of newcomers; she suspects many will spike first and then burn out, because they lack the institutional knowledge and resources of mature organizations, and it is still unclear how many can succeed long term.
What enterprises should change is the procurement list, not buy another AI box
The above are all researcher judgments and data cited in the report, not a verified industry-wide census. Claims on leak sites, institution-deduplicated statistics, and actual intrusions are three different measures; mixing them together to discuss a surge in attacks can easily distort the picture. What can truly inform enterprise decisions is that the number of attacks is rising while the technical sophistication of individual attacks is not necessarily rising: phishing emails are more natural, negotiation scripts are more skilled, and there are more new groups—meaning the weight of identity and email entry points, privilege reduction, offline backups, and recovery drills should increase, rather than betting the entire budget on detection large models.
My judgment is that domestic enterprises have two exposure surfaces in this round of change. First, overseas business: the increase Kaiser mentioned mainly occurs outside the United States, and cross-language, cross-jurisdiction branches often share one set of accounts and approval processes, which is exactly where AI-generated content is easiest to break through. Second, the payout conditions of ransomware insurance and cross-border data compliance: lone operators prefer pure data extortion, and negotiation and disclosure must satisfy both domestic and local regulators, so the playbook cannot simply copy the U.S. template. These two points are inferences based on the report, not conclusions given by Wired.
AI here is leverage, not the protagonist
The reason researchers give for why fully automated AI ransomware has not yet appeared is not mysterious: the existing criminal business model is already very profitable, so there is no need to rewrite it. The reverse also holds: once the cost of a certain link is pushed to a critical point, the rewrite will happen quickly. Liska said groups are testing AI extensively, leaving room in his tone, and Kaiser also said it is still unclear whether new groups can survive long term.
Reading this as AI having created super hackers is inaccurate; a closer-to-fact statement is that AI has broken a business that originally required organization, language skills, and a toolchain into scattered steps that individuals can also start. The cost structure on the defense side has therefore changed: in the past, detection and blocking were based on the characteristics of a few major groups; now defenders face more opponents, with shorter lifespans and less standardized behavior. For security vendors, what can be sold is no longer just a threat intelligence database, but the ability to connect identity, endpoints, backups, and emergency drills into a process; for buyers, the acceptance standard should change from how many sample types were blocked to how quickly business can be restored after data is encrypted or stolen.
Today's Quote
"What a lower technical barrier first changes is often not the strong, but how many people can enter the field. After AI lowers the barrier to the ransomware business, defenders no longer need to count a few groups, but how many entry points they have."
Follow [De Gaohang Zhiqinglang], three hardcore tech interpretations a day, to understand the industry and national destiny behind technology.