Ransomware had a record quarter, and a new survey says artificial intelligence played a role in most attacks. The headline number is 84 percent. The catch is that the number measures what victims believed, not what investigators proved. The latest telling arrived on 14 November 2026, when Wired published a feature titled “Beware the New AI-Fueled Ransomware Boom”. The evidence behind the boom narrative comes from five separate reports published between 14 October and 3 November, plus two earlier cases the boom narrative skips. AI has genuinely operated inside victim networks. But the numbers behind the boom are still perception and volume, not proof of a surge.
That gap matters. Ransomware volume is genuinely up, and fear around AI is genuinely widespread. But the hard evidence that AI drives the surge is thin: one survey of victims, two cases where AI did real intrusion work, one model-run extortion operation, and a forecast that points to 2028.
What the 84% figure actually means
The 84 percent figure comes from the Semperis 2026 Ransomware Risk Report, published on 14 October 2026. Semperis polled 1,200 IT and security professionals across nine countries, with fieldwork run by Censuswide in the first half of 2026, according to Virtualization Review. The sample covered only organisations victimised by ransomware gangs in the previous 12 months, so the results describe the experience of victims.
Within that group, 84 percent said attackers exploited AI in some way. That wording is the whole story: respondents said they believed AI was used. It is survey perception, not a forensic measurement, and it should never appear without that caveat. The same survey found that 44 percent said attackers exploited the victim’s own AI agents or systems, while a separate 40 percent reported AI based attacks.
The report also says that 87 percent of surveyed attacks targeted identity systems such as Active Directory, Entra ID and Okta, and that nearly 75 percent of successful attacks targeted or exploited identity infrastructure. The report puts AI involvement at 84 percent, and 87 percent is the identity figure. The survey also found that 69 percent of respondents had been hit more than once in the past year, and that 40 percent of attacks included psychological or physical threats against employees or their families, with attackers following through in 53 percent of those cases.
Two cases the boom narrative skips
In July 2026, Sysdig documented JADEPUFFER, what it assesses as the first ransomware operation driven end to end by a large language model. The agent exploited a Langflow vulnerability, stole credentials, moved laterally to a production database, encrypted 1,342 configuration records and left a ransom note, adapting to failures without human direction at each step. A human provisioned the infrastructure and chose the victim, so it was not fully autonomous, but the model ran the operation itself.
The second case is Aur0ra. Sophos reported on 29 October 2026 that the GOLD OGRE group used AI agents orchestrated by Cursor’s Composer model and models from Anthropic, xAI and Moonshot to conduct intrusions. The agents ran network and Active Directory discovery, executed commands, moved laterally and attempted credential attacks, reaching domain administrator access in some environments, largely without human input at each step. Three victims later appeared on Aur0ra’s leak site. Humans kept control of the initial access and of the ransomware deployment itself.
The volume boom is real, and it says nothing about AI
The second piece of evidence is a volume record. NordLayer Intelligence recorded 2,894 ransomware incidents in the third quarter of 2026, up almost 50 percent year on year from 1,933 in the third quarter of 2025, and up 8.6 percent from 2,664 in the second quarter of 2026. NordLayer calls it a record high, according to Infosecurity Magazine.
Those counts come from monitoring more than 200 leak sites and the dark web, and they count victims that groups posted or claimed, not every attack and not confirmed breaches. The Gentlemen posted 383 victims and Qilin posted 363, together 746 victim listings, about a quarter of the total. The United States remained the top target, with 805 victim listings, according to NordLayer.
The 50 percent rise is a rise in total listings, not in AI driven attacks. NordLayer does not say how many of the 2,894 listings involved AI, though its report notes Qilin uses AI for personalised phishing campaigns and to generate scripts that disable backups.
The case where AI showed up in the command stream
The third piece of evidence is the only case in the set where the AI behaviour was visible forensically, in the command stream of a live intrusion. Bitdefender published research on 3 November 2026 into a 3AM affiliate intrusion into a UK school estate of 47 systems, including about 21 servers and 8 domain controllers. Hands on activity ran from 10 to 19 September 2026, and about 37 machines were encrypted on the morning of 19 September, between 05:07 and 05:33 UTC.
The attacker stole backup credentials, including credentials from three Veeam backup servers, created two backdoor domain admin accounts called “library” and “svc_bkpmon”, established two way trusts between three previously separate Active Directory domains, and deployed 3AM’s Rust based encryptor. How the attacker first got in was not established. Bitdefender describes the operation as a hybrid: a human operator appeared to choose the goals, while something behaving like an agentic AI loop handled repetitive command generation and retries, with numbered script revisions and 40- to 101-second retry cycles, according to Bitdefender.
Bitdefender did not prove that an autonomous system ran the attack. It could not identify which model was used, where it ran, or whether a person was pasting output into a chatbot, because it never examined the attacker machine. Bitdefender’s own caveat is blunt: this is “one case, an anecdote, and not a trend”, and it found nothing novel in the tooling. It expects hybrid attacks to become more common in 2027 because the approach is cheap and pragmatic. As of 3 November, no leak site listing linked to the attack had surfaced, and the victim organisation has not been named.
What the 2028 forecast does and does not say
The fourth piece of evidence is a forecast. The UK National Cyber Security Centre published its State of the Cyber Nation report on 3 November 2026, warning that AI powered malware is “highly likely” to be used in a ransomware attack on Britain’s critical national infrastructure by 2028, according to the NCSC and Reuters. A forecast for 2028 is a warning about what may come, not an attack.
The AI label some gangs wear
The fifth piece of evidence is a marketing claim. Storm, a ransomware operation that emerged in August 2026, brands itself as the “first AI-powered, quantum-resistant ransomware”. ReliaQuest’s third quarter 2026 report records Storm listing 71 victims over 20 days in August and September, more than DragonForce at 69 and LockBit at 62.
Check Point researchers found no evidence that AI supports Storm’s intrusion, encryption, extortion or affiliate workflows. The AI branding appears to be a recruitment and marketing tool rather than a technical reality, and Storm’s own count of more than 50 victims in six weeks is the group’s claim, not an independently verified figure.
Seven reports, one direction
TechReview analysis: the boom narrative is a pile up of separate evidence, not one new event. None of the pieces is connected to any other as a single campaign or incident.
| Date | Evidence | What it proves about AI |
|---|---|---|
| 1 Jul 2026 | Sysdig JADEPUFFER | Model ran an extortion operation end to end; human set it up |
| 14 Oct 2026 | Semperis survey, 1,200 victims | Victims believed AI was used |
| 21 Oct 2026 | NordLayer Q3 count, 2,894 listings | Nothing about AI; volume only |
| 29 Oct 2026 | Sophos Aur0ra research, Cursor and Claude agents | AI agents ran intrusions; domain admin in some environments |
| 2 Nov 2026 | Check Point Storm research | Marketing claim; no evidence AI supports the operation |
| 3 Nov 2026 | Bitdefender 3AM case and NCSC forecast | One hybrid case, plus a 2028 forecast |
| 14 Nov 2026 | Wired feature | Synthesis, not new data |
The same counts support a second calculation. The third quarter of 2026 had 92 days, so 2,894 listings works out at about 31.5 leak site listings per day. The third quarter of 2025 had 1,933 listings over the same 92 days, about 21.0 per day. That is roughly 10 more victim listings every single day than a year ago. The boom in hard numbers is a volume boom; the AI part is mostly perception plus one anecdote.
What to watch next
Three things will settle whether the AI part of this story is real. First, the fourth quarter NordLayer count, which will show whether the record pace held after September. Second, whether any forensic case after 3AM shows AI doing more than generating retry commands, and whether researchers can name the model and the loop. Third, whether Storm or any other gang produces evidence behind its AI branding. Bitdefender expects hybrid attacks to grow in 2027, and the NCSC forecast points to 2028. Until then, the honest summary is this: ransomware is at a record volume, victims believe AI is involved, and AI has genuinely done intrusion work, but the numbers are perception and volume, and no attack has yet run without a human at the start or the end.
