IBM's 2026 breach study finds no proper AI access controls in 92% of AI-related breaches
IBM and Ponemon put the average data breach at a record $4.99 million, up 12%, in a study of 602 organizations. In Varonis's account, 92% of AI-related breaches lacked proper AI access controls, though AI-driven attacks also added about $1 million each.
The Watch · Security desk
Key figures: AI governance was absent at 68% of breached organizations this year, up from 63% last year. 40% of organizations reported using access controls on AI models and data at all.
| Figure | Value | Note | Claim |
|---|---|---|---|
| Breached organizations without AI governance, this year | 68% | 10 | |
| Breached orgs, last year | 63% | 10 | |
| Organizations using access controls on AI models and data | 40% | as reported by organizations | 7 |
What happened
- The study drew on interviews with 3,558 security and C-suite leaders about breaches between March 2025 and February 2026.
- Shadow AI incidents rose to 43% from 20% and averaged $5.39 million, against $4.63 million the year before.
- Inversion attacks averaged $6.07 million and prompt injection attacks $5.89 million, the costliest incidents in the study.
Why it matters
- contradiction Varonis names access gaps as the real driver of AI breach costs, yet the same summary gives AI-driven attacks a separate cost line.
- cost Extensive AI and automation in the security lifecycle cut breach costs by $1.93 million, about 1.9 times what AI-driven attacks add.
- decision Going without approval gates for AI tools costs money: shadow AI incidents ran $0.40 million above the overall average breach cost.
- exposure An AI assistant connected to broad permissions hands them to whoever compromises it; Varonis's SearchLeak research showed a Copilot exploit inheriting the victim's full permissions.
In its summary of the study, Varonis says the 92% figure makes access and governance gaps the real driver of AI breach costs [21]. The figure itself is a prevalence: 92% of AI-related breaches involved no proper AI access controls [6]. The summary does not include a cost comparison between AI breaches with those controls and those without, and it does not tie the 12% rise in the average to AI [3].
AI also has its own cost line, as an attacker. AI-driven attacks rose 56%, and more than one in four organizations hit by a malicious attack said it came from AI [4]. Deepfake impersonation and AI-generated malware were behind most of that volume and added an average of $1 million per breach [5]. That is about 20% of the $4.99 million average [16].
In our view the data support a narrower claim than the one Varonis makes. Missing AI access controls are near-universal in AI-related breaches, and AI-driven attacks add cost on top of them.
Shadow AI gives the governance argument its clearest dollar figure. Those incidents averaged $5.39 million against $4.63 million a year earlier, a rise of $0.76 million, or 16% [17]. Among organizations with any AI policy, roughly one in three required strict approval before AI could be deployed, and 19% said their governance and security teams coordinate [11].
The environment around the model is where the breaches started. Among organizations with an AI-related breach, the root cause was rarely the model and usually compromised APIs, connected applications or cloud misconfigurations [8]. Varonis says its guidance on least privilege for AI agents is built to close that gap [14].
Identity and access management is reported as one of the most effective cost reducers across breaches overall, yet only 40% of organizations reported using access controls on AI models and data at all [20][7]. Varonis says the problem is execution [20].
What to watch
- Next year's edition: whether absent AI governance at breached organizations climbs past 68%.
- IBM's own report text, for any cost split between AI breaches with and without access controls.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence58
- Adoption40
- Hype gap+20
- Incentives70
- Confidence60
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
IBM's Cost of a Data Breach Report 2026, researched independently by Ponemon Institute, studied 602 organizations across 17 industries.
- [2]
The study drew on interviews with 3,558 security and C-suite leaders with firsthand knowledge of a breach at their organization between March 2025 and February 2026.
- [3]
The global average cost of a data breach reached a record $4.99 million, up 12% from last year.
- [4]
AI-driven attacks increased by 56% since last year's report, and more than one in four organizations that experienced a malicious attack said it resulted from AI.
- [5]
Deepfake impersonation and AI-generated malware were behind most of the AI-driven attack volume and added an average of $1 million per breach.
- [6]
92% of AI-related breaches involved no proper AI access controls.
- [7]
Only 40% of organizations reported using access controls on AI models and data at all.
- [8]
Among organizations that reported an AI-related breach, the root cause was rarely the model itself and was instead the surrounding environment: compromised APIs, connected applications, and cloud misconfigurations.
- [9]
The costliest security incidents were inversion attacks, averaging $6.07 million, and prompt injection attacks, averaging $5.89 million. Varonis says these are attacks on trust in the system itself, which pushes remediation into governance and disclosure.
- [10]
AI governance, covering access controls for models, agents and applications plus monitoring for data leaks, was absent at 68% of breached organizations this year, up from 63% last year.
- [11]
Only about a third of organizations with any AI policy had strict approval processes for deploying AI, and just 19% reported that governance and security teams coordinate with each other.
- [12]
Security incidents involving shadow AI more than doubled to 43% this year from 20% last year, and averaged $5.39 million compared with $4.63 million last year.
- [13]
Organizations that applied AI and automation extensively across their security lifecycle cut breach costs by $1.93 million.
- [14]
Varonis says its guidance on least privilege for AI agents is built to close the access control gap.
- [15]
Varonis Threat Labs' SearchLeak research found that a Copilot exploit let an attacker inherit the victim's full permissions.
- [16]
The $1 million average added per AI-driven breach equals about 20% of the $4.99 million average breach cost.
- [17]
Shadow AI incident cost rose $0.76 million, about 16%, from $4.63 million to $5.39 million.
- [18]
Shadow AI incidents averaged $0.40 million above the $4.99 million overall average breach cost.
- [19]
The $1.93 million saving from extensive AI and automation in security is about 1.9 times the $1 million AI-driven attacks add.
- [20]
Identity and access management is reported as one of the most effective cost reducers across breaches overall, and the AI access control gap persists despite that, which Varonis says suggests the issue has more to do with execution than awareness.
- [21]
Varonis states that the 92% finding makes access and governance gaps the real driver of AI breach costs.
Get this ledger inside ChatGPT. Ask about any of our stories and the answer brings its figure, each claim’s verdict and the sources. Free, no account.
Sources
1 independent publisher whose own reporting we read for this story.
- varonis.comIBM’s Cost of a Data Breach Report 2026: AI Is Now the Attacker and the Target
1 article · October 14, 2026
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.