Estimated reading time: 7 minutes
Both numbers sit inside IBM’s 2026 Cost of a Data Breach Report. The gap between them may be the most useful finding in the study.
As follow-on research to the IBM data breach report, Ponemon Institute studied 602 breached organizations between March 2025 and February 2026. Nearly two-thirds said they planned to spend more on security. Researchers went back in May 2026. By then, Anthropic had announced Claude Mythos Preview. The model had found thousands of high-severity vulnerabilities across major operating systems and browsers. Among firms aware of that news, 85% said they would increase spending. Cyber insurance underwriters take note.
Fear of the hypothetical beats the memory of the actual.
Breach Costs Set A Record
The global average breach cost reached $4.99 million. That marks a 12% rise and an all-time high. The trend line has swung hard. The average ran to $4.88 million in 2024. It fell to $4.44 million last year.
Detection, escalation and lost business drove 63% of the total. Those categories cover forensics, crisis management, downtime and customer churn. Regulatory fines carry less weight than most boards assume.
US breaches averaged $11.5 million, up from $10.22 million. That runs to more than twice the global average.
Response times moved the wrong way. Organizations took 247 days to identify and contain a breach. That reversed a five-year decline. Breaches running past 200 days cost $5.65 million. Shorter ones cost $4.32 million.
Financial Services Moved Up The Table
Healthcare held the top spot for a 13th year at $6.64 million. That figure fell 10.5% from $7.42 million.
The sector’s exposure has been shifting toward its suppliers. A separate 2026 survey found 85% of practices suffered a third-party disruption in the past year.
Financial services came second at $6.29 million, up from $5.56 million. Energy reached $5.24 million.
IBM says AI-driven attacks concentrated on critical infrastructure. Those sectors absorbed 62% of AI-driven breaches in the study, with financial services and energy carrying most of that share.
The initially published report muddies the figure. Page 35 states that financial services and energy alone experienced a combined 62%. In an email exchange seeking to clarify, IBM told CINI that wording is wrong. The 62% covers all critical infrastructure sectors, not those two industries.
One gap remains. IBM does not publish which of its 17 categories it counts as critical infrastructure. Without that list, readers cannot measure how far financial services and energy run above their share of the sample.
AI Attacks Added A Million
More than one in four organizations hit by a malicious attack said it was AI-driven. That represents a 56% rise. Another 11% could not tell either way.
Those breaches averaged $6.04 million. Malicious attacks without AI averaged $5.03 million.
Read the comparison carefully. IBM’s press release describes the $1 million gap as sitting above the $4.99 million global average. The report itself compares it to the malicious average. The two baselines happen to sit close together.
Deepfake and impersonation attacks drove 45% of AI-driven incidents. AI-enabled malware accounted for 19%. Carriers have started answering with policy language. BOXX added affirmative AI and deepfake cover to its commercial product this month.
“What’s changing is the economics of cyberattacks,” said Suja Viswesan, VP of IBM Security Software.
Attacks on AI systems grew too. AI-related breaches reached 21%, up from 13%. Among those firms, 92% lacked proper AI access controls. Shadow AI incidents more than doubled to 43%.
What The Controls Actually Buy
IBM’s headline savings figure is $1.93 million. Firms using AI and automation extensively averaged $4 million per breach. Firms using neither averaged $5.93 million.
Figure 32 tells a smaller story. It prices 30 factors against the global average. A DevSecOps approach leads at $253,805. Identity and access management follows at $225,622. Encryption saves $213,478. The worst amplifier is a supply chain breach at $227,250. Security system complexity adds $208,265.
The full spread runs about $481,000. That sits under a tenth of the average breach. (CINI calculation)
Both findings appear in the same report. They measure different things. Figure 32 isolates each factor against the average. The $1.93 million figure compares two groups of companies. That second comparison captures budget, headcount, and security maturity alongside the tools.
Underwriters price control checklists. IBM’s own data values each item on that checklist at roughly 5% of a breach.
Joshua Brown, CISO at Spektrum Labs, made a related point on the CINI podcast this month. “You can be a hundred percent compliant and still be breached,” he said.
Agents Went Where The Risk Was Not
Half of breached organizations deployed AI agents in the security operations center. Most aimed them at threat hunting, at 56%. Automated response and containment followed at 54%. Only 18% applied agents to vulnerability scanning and management. Frontier models target exactly that gap.
Vendors are moving to close it. Command Zero previewed Throughline ahead of Black Hat USA 2026, a capability that reopens closed cases when new evidence lands.
The follow-on research shows firms noticing. Some 74% rethought where they deploy agents. Planned use in vulnerability management rose to 37%.
Mark Hughes, IBM’s global managing partner for cybersecurity services, told Infosecurity Magazine that “attackers can now execute attacks in minutes rather than days with advanced frontier models.”
Why It Matters For Cyber Insurance Underwriters
Three numbers deserve a place in renewal conversations. The first is 247 days. Longer lifecycles cost $1.33 million more. Any control that shortens detection earns its price.
The second is 92%. Nearly every organization breached through its AI systems lacked access controls. Only 40% apply access controls to AI models and data at all. That question belongs on applications now.
The third is 62%. Critical infrastructure absorbed that share of AI-driven attacks, with financial services and energy carrying most of it. IBM does not say which sectors it counts as critical infrastructure, so the concentration cannot yet be priced against sample share.
Key Findings
- $4.99M Global average breach cost, up 12% to a record
- $6.29M Financial services average, second only to healthcare
- $6.04M Average AI-driven malicious breach, against $5.03M without AI
- 56% Rise in AI-driven attacks year over year
- 92% Share of AI-breached firms that lacked AI access controls
- 43% Share of incidents involving shadow AI, more than double last year
- 247 days Mean time to identify and contain, reversing a five-year decline
- $253,805 Value of the single best cost-reducing factor, a DevSecOps approach
- 18% Share of AI agents applied to vulnerability management
- 85% Share of frontier-AI-aware firms planning to raise security spending, against 64% after a breach
Methodology
Ponemon Institute conducted the research. IBM sponsored, analyzed, and published it. Researchers studied 602 organizations across 17 industries and 16 countries. Breaches ran from 2,590 to 115,380 compromised records between March 2025 and February 2026. The team ran 3,558 interviews.
A follow-on study ran in May 2026. Of the original 602 organizations, 456 responded. Some 78% of those said they were aware of frontier AI model reports.
IBM states that its sample is nonstatistical. Margins of error and confidence intervals do not apply. The sampling frame skews toward organizations with mature security programs. Industry cells vary widely in size. Financial services made up 13% of the panel. Healthcare made up 2%.
FAQ – IBM Cost of a Data Breach 2026
IBM puts the global average at $4.99 million. That marks a 12% rise over last year and a record high.
AI-driven malicious breaches averaged $6.04 million. Malicious breaches without AI averaged $5.03 million. The gap is about $1 million.
Healthcare, at $6.64 million, for a 13th consecutive year. That figure fell 10.5% from last year. Financial services came second at $6.29 million.
Shadow AI describes unapproved AI tools used by employees. Incidents more than doubled to 43% this year. Roughly one in five led to a regulatory fine.
No. IBM states the sample is nonstatistical. Margins of error and confidence intervals do not apply. Industry averages rest on small cells.
Editor’s Note
After reviewing the report, CINI identified two errors in IBM’s published findings and asked IBM to confirm them. The initial report stated on page 13 that average breach costs increased in all industries; IBM confirmed the figure should read that costs increased for all other industries besides healthcare, which fell 10.5%. The report stated on page 35 that financial services and energy alone experienced a combined 62% of AI-driven breaches; IBM confirmed the 62% covers all critical infrastructure sectors, with financial services and energy making up the majority. IBM said it would correct these elements. Our article has been updated to clarify these elements. Updated July 29, 2026.
Related Cyber Insurance Posts
- Arctic Wolf: A Third Party Made The Call In Half Of Ransom Payments
- Agentic AI: 98% Have Already Had An Incident. Few Can Prove What Happened(Opens in a new browser tab)
- IBM 2025 Cybersecurity Report: Credential Theft Skyrockets(Opens in a new browser tab)