AI breach risk is forcing a familiar corporate problem into the open: the technology team is often asked to manage a business risk that it lacks the authority to govern. CIO.com points to findings that show AI-related incidents, shadow use of AI tools and weak basic controls are rising together.
The response cannot stop at more security tools. Leaders need clear ownership, tested safeguards and enough capacity to act before an incident becomes a crisis.
- IBM’s 2026 report, based on 602 breached organizations, found one in four malicious breaches involved AI.
- Those AI-enabled breaches averaged $6 million, compared with a global average of $4.99 million.
- Shadow AI appeared in 43% of breached organizations, up from 20% the previous year.
- Breaches involving AI systems represented 21% of incidents.
- IBM reported an average of 247 days to identify and contain a breach.
===
What the New Breach Numbers Say
The figures reported by CIO.com describe more than a new attack technique. They show that trusted tools, connected infrastructure and unauthorized use are pulling risk across organizational boundaries.
Compromised APIs, plug-ins and misconfigured cloud workloads were highlighted as common entry points in incidents involving AI systems. The model itself was not always the weak point. The surrounding infrastructure was.
At the same time, 53% of breached organizations had not encrypted sensitive data both at rest and in transit. Advanced concerns were therefore sitting beside basic gaps. Security teams cannot assume that AI budgets alone address either layer.
The response to AI breach risk needs to cover the full route into the organization: employee use, vendor access, software interfaces, cloud configuration, data protection and executive decision-making.
The Perimeter Problem Comes Back
The Merck & Co. Case described by CIO.com shows why familiar controls can fail. NotPetya entered in 2017 through compromised accounting software supplied by a Ukrainian vendor. It spread to about 40,000 computers, and Merck placed the damage at more than $1.4 billion. Insurers later disputed coverage because of policy terms connected to an attributed act of war. A confidential settlement was reached in January 2024.
The useful lesson is not that every organization needs to predict the next malware family. It is that trusted traffic and supplier relationships can carry serious threats through the doors created for legitimate business.
Johnson Controls offers another example. Ransomware entered through its Asia offices in 2023, then spread across the network. The company confirmed data theft and more than $27 million in costs, net of insurance.
Perimeter controls still matter. They simply cannot carry the entire risk model.
Governance Is the Control That Multiplies
CIO.com also cites a survey of 113 CISOs from IANS and Artico Search. Security leaders’ confidence about the next two years tracked leadership understanding, clear ownership, budget authority and staffing more closely than the maturity of an AI security program alone.
That is an organizational finding, and it is awkward for companies that place cyber risk under IT while asking IT to monitor the executive decisions creating that risk. AI expands the number of people and systems generating risk, from procurement teams selecting tools to business units introducing AI into customer processes.
A practical governance structure should identify:
- An executive accountable for AI-related business risk.
- A named owner for each approved AI tool and use case.
- Security and privacy requirements attached to procurement.
- A process for discovering shadow AI.
- Escalation rules for incidents affecting customers, operations or sensitive data.
- Tested exit and recovery procedures for critical vendors.
The point is not to create another committee. It is to put decisions where authority and information meet.
A Worked Example for Risk Review
Say a sales team starts using an unapproved AI writing tool to summarize customer calls. The security team notices the tool through a cloud audit, but the sales leader treats it as a productivity choice and the IT team cannot approve or prohibit it.
Under an AI breach risk model, the product owner of that workflow should clarify the data involved, permitted uses and review rights. Security should test access controls and data handling. Procurement should establish whether the tool can remain in the stack. An executive should own the residual risk if the workflow continues.
That is not bureaucracy. It makes the speed of the business decision visible and gives security a mandate before customer data becomes an incident.
What to Watch Next
The most important signal is organizational, not technical. Watch whether AI ownership moves closer to the leaders deciding where AI is used, whether security receives authority beyond advice, and whether basic controls are being completed before organizations buy more sophisticated defenses.
IBM’s reported cost figures make the problem expensive. The governance gap may make it persistent. Organizations that treat AI breach risk as an IT workstream will continue reacting after trusted paths fail. Treating it as enterprise risk gives controls an owner, a budget and a business mandate.
Key takeaways
- AI-related incidents expose weaknesses in APIs, plug-ins, cloud configuration and basic data protection.
- Shadow AI requires discovery, governance and clear escalation paths.
- Security leadership gains more confidence when ownership, authority and staffing are clear.
- Executive decisions about AI use must remain linked to enterprise risk accountability.
Frequently asked questions
What does AI breach risk include beyond attacks on AI models?
It includes compromised APIs, plug-ins, misconfigured cloud workloads, shadow AI tools and weaknesses in basic data protection.
Why do trusted vendors and perimeter controls remain important?
Legitimate supplier access can provide a route into an organization, as illustrated by the NotPetya case involving Merck’s accounting software supplier.
How should a company govern unauthorized AI use?
It should identify the tools and workflows involved, assess data and access risks, set approval requirements and assign accountable owners.
What did the IANS survey of 113 CISOs indicate?
Confidence about the next two years tracked leadership understanding, clear ownership, budget authority and staffing more closely than program maturity alone.