AI is changing how cyberattacks operate, but the harder problem may be organizational. According to cio.com, one in four malicious breaches in IBM’s 2026 Cost of a Data Breach report was AI-enabled.
That makes AI security governance a leadership issue. Who owns the risk, who can act on it, and does the accountable person have enough authority to influence budgets and business decisions?
- The 2017 NotPetya attack on Merck caused more than $1.4 billion in damage.
- Breaches taking over 200 days to find and contain cost about $1.3 million more.
- Some 64% of CISOs report to IT leadership, while only 10% influence financial decisions.
- Only a quarter of CISOs said board discussions of cyber risk exceed 30 minutes.
- Among organizations with an AI-related incident, 92% lacked proper AI access controls.
How the responsibility gap developed
The reference traces today’s governance problem to the 2017 NotPetya attack on Merck. The malware entered through trusted vendor software and routine traffic, exploiting the paths companies had built their defenses to permit. Merck later estimated the damage at more than $1.4 billion.
The dispute that followed between Merck and its insurers changed expectations. Insurers began pressing for stronger controls, clearer ownership and evidence that basic protections worked.
AI now adds speed and scale to the threat. Deepfake executives, conversational software and credential-holding agents target trust before technology. But the internal operating model often remains unchanged: security teams carry responsibility for outcomes while reporting to leaders who oversee the systems being assessed.
That is not simply a tooling deficit. It is an authority problem.
What changed now
Every AI tool adopted by a business is also a security design decision. Yet, according to cio.com, only 19% of organizations report coordination between their AI governance and security teams.
That separation creates predictable friction. One group approves a tool, another deploys it, and security learns about its permissions and risks later. Each function may have completed its task, while nobody owns the decision end to end.
AI agents make this harder to ignore because they can act using identities and permissions. The reference recommends giving every non-human identity a named owner, minimum access, monitoring and a kill switch. Among organizations that experienced an AI-related incident, 92% lacked proper AI access controls.
Boards also need a clearer measure than completed controls. Detection, containment and recovery times should be tested rather than estimated. According to the reference, the difference between breaches resolved within 200 days and slower cases is worth about $1.3 million.
Our analysis: authority must match accountability
A CISO reporting to the CIO may appear tidy, but it can create self-review. The security leader must assess systems overseen by the executive above them. More importantly, the arrangement can limit their ability to challenge spending or stop risky work.
A different reporting line is not automatically better. The reference notes that Vertex Pharmaceuticals places its CISO under the Chief Scientific Officer, while Equifax established a direct line to the CEO and board after its 2017 breach. The lesson is to choose deliberately, preserve technical credibility and ensure access to decision-makers.
Elevating the CISO also should not create a remote-control role. A security leader may understand threats without fully understanding research, manufacturing, claims processing or customer operations. A standing risk council can bring security, business leaders, legal, finance, technology and operational owners into the same decision before controls are imposed.
This matters for delivery teams because security decisions made in isolation tend to be worked around. Decisions shaped with the affected teams are more likely to be followed.
What this means for agile teams
Say a Scrum team is building an AI-assisted support tool. Before release, the team should agree on who owns the tool’s credentials, what data it can access, how its activity will be monitored and who can disable it. If the tool introduces a policy conflict, the product owner raises it during planning while the security lead evaluates access and containment.
That turns a broad policy into work the team can plan, review and own. Governance becomes part of delivery rather than a final gate.
The same discipline applies to responsibility across the organization. Technology teams execute technical controls. Business leaders accept or reject risk in the context of operations. The board asks trade-off questions and confirms that accountability comes with authority.
Who should care, and what happens next
CISOs need a reporting structure that allows them to challenge decisions, not merely report them. Boards need evidence about response times and investment trade-offs. Business leaders must treat AI risk as part of operational risk. Technology and delivery teams need requirements that arrive early enough to shape the work.
A practical reset can proceed in four steps. Establish one owner for each AI adoption decision. Define the CISO’s authority and escalation path. Create a standing forum for security and business leaders. Assign owners, least-privilege access, monitoring and deactivation to every non-human identity.
What to watch next
Watch whether AI governance becomes a standing executive responsibility or remains buried inside tooling reviews. Useful signals include named owners for AI agents, participation by business leaders in risk decisions and response-time targets tested through exercises.
The decisive sign will be a board discussion focused on budget, operations and risk rather than regulatory status alone.
Our stance is clear: AI security governance cannot belong only to the teams that administer AI. It needs an accountable owner, real authority and shared participation from the people whose work will be affected. Everything else is an operating detail.
Key takeaways
- Give the CISO a deliberate reporting line and enough authority to influence business and budget decisions.
- Make AI governance and cybersecurity joint responsibilities with one accountable owner per decision.
- Assign owners, least-privilege access, monitoring and rapid deactivation to non-human identities.
- Measure tested detection, containment and recovery times, then present the results to the board.
- Bring security into planning early so controls fit the work instead of blocking it at release.
Frequently asked questions
Why does a CISO’s reporting line matter for AI security?
A CISO reporting to IT can be positioned to assess systems overseen by the same leader. A more independent line can strengthen scrutiny, provided the CISO retains technical credibility and collaborates with business leaders.
What controls should organizations apply to AI agents?
Each agent should have a named owner, only the access it needs, monitoring and a kill switch. These controls reduce the risk of excessive or untracked permissions.
How should boards evaluate cybersecurity performance?
Boards can ask for tested detection, containment and recovery times rather than relying on checklists. They should also examine whether cybersecurity is treated as a budget and business-risk decision.
Is AI security solely the CISO’s responsibility?
No. The CISO should own the risk perspective and participate in decisions, but legal, finance, technology, business owners and operational leaders must help shape controls they will be expected to follow.
What should an agile team do before deploying an AI-enabled tool?
The team should clarify ownership, permissions, monitoring and deactivation before release. Risks and exceptions should be handled through normal planning and review, with security involved early.