
On July 29, 2026, IBM published its annual Cost of a Data Breach report. The key finding: one in every four malicious breaches is already AI-enabled — a 56% increase over last year. These breaches cost $6 million on average, about $1 million more than the global average of $4.99 million.
The Economics of the Attack Have Changed
The important point in the report is not the number but the asymmetry. Attacks are becoming faster and cheaper to launch, while the breach itself is becoming more expensive to detect and remediate. Most AI-enabled attacks are deepfake impersonation and AI-based malware.
What is changing is the economics of cyberattacks. AI is making attacks faster and cheaper, while breaches continue to get more expensive. When an organization has a prolonged gap between detection and remediation, that imbalance shows up directly in the cost of the breach.
The Weak Link Is Not the Model
More than 20% of organizations reported a breach that targeted AI models or applications. But the most common causes were not in the model itself — they were in the environment around it: compromised APIs, applications and plugins (27%), and cloud misconfigurations that affected AI workloads (27%).
This is a point worth pausing on. The public discussion about AI security focuses on the model — will it leak data, can it be tricked. The data shows that attackers get in through the same old, familiar weaknesses: permissions, configuration, interfaces.
Additional Findings
- Most AI-enabled attacks targeted critical infrastructure (62%). A breach in financial services cost $6.3 million on average, and in energy $5.2 million.
- Only 37% of breached organizations encrypt sensitive data both at rest and in transit. Only 34% know which cryptographic assets they have at all.
- Ransomware incidents rose from 34% to 39%. Attackers have shifted from operational disruption to pressure on reputation (41%) and on employee data (35%).
- Organizations that used AI and automation in the security operations center saved almost $2 million on average — and still, one in four has not adopted these tools.
- More than 50% run agents for threat detection and containment, but only 18% run them for vulnerability management — exactly where attackers are fastest.
The report was conducted by the Ponemon Institute and is based on breaches at 602 organizations worldwide between March 2025 and February 2026.
What This Means for Your Organization
- The finding about APIs and cloud misconfigurations is the good news: these are exactly the things a managed IT operation is supposed to cover. You do not need dedicated AI expertise to close most of the gap.
- Check what your actual encryption rate is — both at rest and in transit. 37% is a worryingly low number, and it is easy to discover you are part of it.
- If you run any AI agents, map their permissions. An agent with overly broad permissions is a superuser that nobody is monitoring.
- The gap between the 50% who run agents for threat detection and the 18% who run them for vulnerability management is an immediate opportunity for improvement.
