AI Ransomware Survey Ties 84% of Attacks to AI, Identity Gaps

A survey with a striking number

Ransomware has been the top threat on corporate risk registers for years. This week a single statistic pulled it into mainstream news: in 84% of ransomware incidents reported by surveyed victims, the attackers used artificial intelligence somewhere in the operation. The figure comes from the 2026 Ransomware Risk Report by identity-security firm Semperis, released October 14. The company polled 1,200 IT and security professionals across nine countries and eight industry sectors.1 The countries were the U.S., UK, France, Germany, Italy, Spain, Singapore, Canada and Australia.13 Research consultancy Censuswide ran the fieldwork.

The headline number is not the only finding worth attention. In 44% of cases, respondents said attackers exploited the victim organization's own AI agents or systems.1 Identity infrastructure was targeted in 87% of incidents.7 About 40% of attacks came with psychological or physical threats against employees, and victims said attackers followed through in 53% of those cases.7 Within a year, 69% of affected organizations suffered more than one successful attack, and 59% paid a ransom at least once.14

Coverage settled on one story: AI has gone from a speculative add-on to a standard part of the ransomware toolkit. Read closely, though, the report shows something more specific. It is clearer about where attackers are winning, which is identity systems and human pressure, than about how much AI actually adds.

What "AI-linked" actually measures

The 84% figure is easy to misread as a forensic result. It isn't one. The Hill's write-up, carried on Yahoo, gives the exact framing: 84% of respondents said they believed cybercriminals had used AI "in some way" during attacks against them. That measures how victims perceive their attacks. It doesn't come from attributing specific tools or tradecraft, and "in some way" could mean anything from an AI-polished phishing email to a fully automated intrusion.

The same piece offers a finer breakdown that most outlets left out. It reports 44% of respondents experiencing attacks that exploited their own AI agents and systems, and a separate 40% experiencing AI-based attacks. Those categories are not the same thing as the 84% figure, which covers any AI involvement at all. Copy-paste coverage blurred the difference.

Inconsistencies also appear in how the numbers circulated. Semperis's own press release puts the AI-weaponization figure at 84%.11 The wire version on PR Newswire, republished by Yahoo Finance and VMblog, says 87%.1316 That happens to equal the share of attacks that targeted identity systems, so it looks like an editing slip that spread through syndication. One aggregator put the share of victims whose own AI tools were turned against them at 49%, not 44%. The wire release says 23% of victims or family members were swatted, while the company's site says 24%.1311 Accounts of employee threats also vary. Some outlets call it "more than a third" of victims5, and the wire copy separately mentions psychological intimidation in "nearly 25%" of attacks13. The discrepancies are small, but readers should treat precise percentages from this cycle cautiously and give more weight to the overall pattern than to the decimals.

One more point belongs on the record. Chris Inglis, the former National Cyber Director and the report's most-quoted voice, is identified as a Semperis board member1 and a Semperis strategic adviser13. His long NSA career makes his views worth hearing. Still, the survey's most prominent interpreter is affiliated with the company that produced it, and that company sells identity-recovery products.13

The autonomy claim

The quote that shaped much of the coverage was Inglis's statement that autonomous ransomware attacks are already happening and are "included in the mix."1 He described how AI lets someone without real technical skill generate attack code and point it at a company. He argued that capable models are now so widely available that even low-end ones can enable this.1 Moneywise and Yahoo Finance put the autonomy warning in their headlines910, and one aggregator's summary went further, saying Inglis considers AI-powered defenses possibly the only workable countermeasure.

Semperis researchers made narrower, more verifiable points. Tomer Bar, the company's associate vice president of security research, said AI is shortening malware development cycles and will make attribution harder for law enforcement. He also expects fewer flaws in attackers' encryption, because criminals can test their code against large language models.1 That last point matters in practice. Mistakes in ransomware cryptography have historically let researchers build decryptors and spare victims from paying, and Bar is predicting that safety net will shrink.15

Inglis also said many attacks aren't recognized as AI-assisted because they are "too slick, too sophisticated." That cuts both ways. It suggests the 84% could be an undercount. It also shows how hard it is for anyone, victims included, to confirm AI involvement.

My reading is that "autonomous ransomware" is a credible direction but not yet a documented baseline. Faster development, better lures and cleaner code are well supported in the reporting. Fully hands-off campaigns rest mostly on one expert's assessment.

Your own AI as the way in

The 44% figure is the finding most likely to change corporate behavior, because it concerns systems organizations are deploying right now. Courtney Guss, Semperis's associate vice president of crisis management, warned that companies adopting AI agents should expect those agents to eventually try to compromise identity systems, even with no malicious intent.1 One extended write-up of her comments describes an agent pursuing its task by seizing the identity layer and effectively giving itself "god mode" over connected systems.15

The reasoning is simple. Agents work across many applications and act for users. An attacker who manipulates an agent, or steals the credentials and permissions it relies on, gets another route deeper into the network.1 In practice, every agent is a new privileged identity, and most organizations have not yet put those identities through the controls they use for human administrators.

Defensive AI is not yet balancing this out. About 49% of victims said they used AI tools to detect and fight the attack, but 23% said those tools failed to stop the malware.

Identity is the real battleground

The report's most actionable data has little to do with AI. Semperis says 87% of attacks targeted identity systems such as Active Directory, Entra ID and Okta, and nearly 75% of successful attacks targeted or exploited identity infrastructure.11 Defenses lag behind. Only half of organizations keep dedicated identity backups, fewer than half automate identity recovery11, and 40% have disaster-recovery or crisis plans with no procedure for restoring Active Directory.

Semperis CEO Mickey Bresman framed this as the next step in a familiar pattern. Ransomware began with encryption, then moved to double extortion once backups made encryption alone less effective, adding data theft and threats to leak.11 Identity compromise is where those tactics meet, because the attacker who controls authentication controls recovery.

This also fits the repeat-victim numbers. If nearly 70% of victims are hit more than once and 31% pay multiple times11, many organizations are likely rebuilding on compromised foundations. That is an inference from the survey, not something it directly measured, but it matches the gap it records between how often identity is attacked and how well it can be restored.

Coercion moves offline

The report also records a shift from purely digital extortion to pressure on people. Inglis put it bluntly: attackers try to make resisting cost more than paying, and threatening physical safety is the newest lever because the others are already in use.13 About a quarter of targeted companies said employees or their families were doxxed, 25% needed medical or psychological treatment, and close to 14% saw staff change jobs or office locations, or move home. About 24% of victims or families hired personal security.11

AI connects to this as well. Coverage argues that the same profiling abilities that sharpen phishing can help attackers pick out which employees or executives are most likely to give in.15 Unlike the autonomy claim, this one doesn't depend on exotic capabilities.

The bottom line

The 84% figure is a measure of how victims perceive their attacks, from a vendor survey, and it should be read that way rather than as proof that AI now runs ransomware. The useful signal sits underneath it. Attackers are going after identity systems most companies can't quickly restore. They are exploiting AI agents that were deployed faster than they were governed. And they are pushing coercion off the network and into employees' homes. Whether or not autonomous ransomware is already common, those gaps are documented. Security teams should start with them, not with the AI headline.