Estimated reading time: 7 minutes
Eighty-seven percent of security teams now use AI in their operations. Sixty-four percent have never rehearsed an AI incident. Eight percent drill for one regularly.
A plan nobody has run is a document, not a capability. Underwriters already know that distinction. It is why applications ask whether backups have been tested, not simply whether they exist. Those figures come from ISACA’s 2026 State of Cybersecurity survey, published on 22 September. The research covers more than 1,800 security professionals worldwide, including 494 in Europe.
What The AI Incident Response Numbers Show
ISACA accounts for the rest. Seventeen percent fold AI incidents into broader cyber exercises. Sixteen percent plan to start.
Playbooks lag in the same pattern. Forty-eight percent of respondents either say their organization holds no AI incident playbook or do not know whether one exists.
Adoption moved faster than the drill. Among teams using AI, 41 percent automate threat detection and response, up from 32 percent in 2025. Forty percent automate routine security tasks, up from 28 percent.
Security staff have also moved into AI governance. Fifty-one percent now help develop, onboard or implement AI systems. That figure stood at 40 percent last year and 29 percent in 2024. Fifty-six percent helped write a policy governing AI use.
Jon Brandt, ISACA senior director for professional practices and innovation, named the gap. Many organizations “have not yet matched that adoption with the response planning” required, he said.
Europe Writes More Policy And Runs Fewer Drills
The European cut covers 494 respondents. It diverges from the global picture in a consistent direction.
| Involved in developing AI governance policy | 60% | 56% |
| Involved in AI implementation | 54% | 51% |
| Conducted no AI-related incident exercises | 71% | 64% |
| Conduct AI exercises regularly | 5% | 8% |
| Expect an attack within a year | 54% | 45% |
| High confidence in detection and response | 39% | 42% |
European respondents report higher involvement in AI governance and implementation than the global sample. They also report a larger share with no AI incident exercises at all.
Treat the regular-exercise figures with care. Europe records 5 percent against 8 percent globally. That gap rests on a base of 494 against a sample above 1,800. A few points there may reflect sampling rather than practice. The 71 against 64 comparison carries more weight.
European respondents also expect more trouble. Fifty-four percent anticipate an attack within a year, against 45 percent globally. Thirty-nine percent report high confidence in their team’s ability to detect and respond, against 42 percent.
Why AI Incident Response Reaches The Application Form
Incident response planning is an underwriting question. Carriers ask about plans, playbooks, and tabletop frequency on applications and ransomware supplements.
Carriers wrote those questions for conventional incidents. An insured can answer yes to holding a tested incident response plan while never having exercised an AI-driven scenario.
ISACA lists what such scenarios cover. Sensitive data exposure through AI systems appears in 24 percent of AI-related exercises. AI-enabled phishing, fraud, or social engineering appears in 23 percent. Misuse of generative AI by employees or insiders appears in 21 percent.
Each maps to an existing coverage trigger. None appears on a standard supplement as its own question.
The Skills Behind The AI Incident Response Gap
ISACA also records what teams say they lack. Forty-five percent name LLM SecOps as a skill gap among cybersecurity professionals. That figure rose 12 points from 2025 and 21 points from 2024.
Thirty-one percent name ML SecOps. The same share names cloud computing, and the same share names data security.
Staffing pressure sits underneath. Fifty-eight percent say their cybersecurity team is understaffed, up from 55 percent last year. Forty-nine percent report open positions. Fifty-five percent struggle to retain qualified professionals.
Stress now leads the reasons people leave, cited by 52 percent, up from 47 percent in 2025. Sixty-eight percent say their role grew more stressful over five years. Seventy-one percent blame the complexity of the threat landscape, up from 63 percent.
Thirty-five percent of teams now lean harder on AI or automation to close technical skills gaps. That share rose 12 points in a year.
So teams reach for more AI partly because they cannot hire. They also name the AI security skill as their fastest-growing gap.
Get The Cyber Insurance News Podcast
Photos, MRIs, X-rays, contracts. Every business decision now rests on digital records, and anyone can alter them in seconds. DigiCert’s Mike Nelson explains how provenance technology tracks where a file came from and every change made to it.
A Second Survey Points The Same Way
Cohesity published its fifth annual Global Cyber Resilience Report on 16 September. Vanson Bourne surveyed 3,200 IT and security decision-makers across 12 countries for that research.
It found that 39 percent say their response and recovery plans comprehensively account for attacks targeting AI systems. Three percent say their plans suit frontier AI threats.
Different population, different sponsor, same direction.
Where Practitioners Say Detection Belongs
ISACA published a practitioner discussion alongside the report. One exchange bears directly on control evidence.
Harshad Kadam is a senior security engineer and founding chapter lead of AI Security Engineers Austin. He argued that organizations are balancing the wrong thing. The incidents that matter happen because “nobody can see what the agent did,” he said.
He placed detection at the network layer and the tool-call boundary. “Detection belongs where its actions are observable,” he said, rather than in a model’s own account of itself.
Saurabh Misra of ManpowerGroup described the same lag from the governance side. Controls, testing, monitoring and accountability “often lag behind adoption,” he said.
Both point at what an underwriter needs. Logged agent behavior produces evidence. A policy document does not.
Get The Cyber Insurance Upload
Subscribe to our weekly newsletter!
What Cyber Insurance Underwriters Can Ask Now
Three questions follow from the data.
- Has the insured run an incident response exercise covering an AI-driven scenario, and when?
- Does an AI incident playbook exist as its own document, separate from the general plan?
- Can the insured produce logs of agent actions rather than records of its AI policy?
None of those requires new wording. Each separates an organization in ISACA’s 8 percent from one in its 64 percent.
Methodology
ISACA surveyed its own professional community rather than a random sample of organizations. Wolters Kluwer TeamMate sponsored the report. The survey reached more than 1,800 professionals globally, with 494 responses from Europe. This is the twelfth annual edition. ISACA will host a webinar on the findings on 20 October.
FAQ – AI Incident Response
What did ISACA find about AI incident response?
Eight percent of organizations run AI-specific response exercises regularly. Sixty-four percent have run none at all. Seventeen percent include AI incidents in broader cyber exercises and sixteen percent plan to start.
How many security teams use AI in their operations?
Eighty-seven percent. Among them, 41 percent automate threat detection and response, up from 32 percent in 2025, and 40 percent automate routine security tasks, up from 28 percent.
How does Europe compare with the global sample?
European respondents report more involvement in AI governance at 60 percent against 56, and a larger share with no AI incident exercises at 71 percent against 64. They also expect more attacks, at 54 percent against 45.
What AI scenarios do organizations exercise?
Sensitive data exposure through AI systems at 24 percent, AI-enabled phishing, fraud or social engineering at 23 percent, and misuse of generative AI by employees or insiders at 21 percent.
Why does this matter for cyber insurance?
Carriers ask about incident response plans and tabletop frequency on applications. An insured can hold a tested plan and still have never exercised an AI-driven scenario, so a yes answer may not describe AI cyber insurance readiness.
Who was surveyed?
More than 1,800 cybersecurity professionals globally, including 494 in Europe. ISACA surveyed its own professional community, and Wolters Kluwer TeamMate sponsored the report.
Related Cyber Insurance Posts
- Shop Early for Cyber Insurance and Don’t Forget Response Exercises: Wall Street Journal (Opens in a new browser tab)
- Cyber Resilience Under Fire: New Data Exposes a Global Confidence Gap(Opens in a new browser tab)
- Cyber Incident Response Plan: Reasons It Fails & How to Fix It(Opens in a new browser tab)
- Agentic AI: 98% Have Already Had An Incident. Few Can Prove What Happened(Opens in a new browser tab)