Intro
96% of CISOs now have formal responsibility for AI governance and risk management. That sounds like progress. It isn't. What actually happened is that organizations looked for the nearest risk-aware leader, pointed at them, and called it a strategy. But AI governance covers legal liability, procurement risk, data ethics, workforce impact, and business model decisions that sit well outside what a CISO has ever been asked to own. Without changing the org chart around them, this assignment is a setup for failure. The organizations getting it right already know that.
The default is broken
When a new risk category shows up, and nobody knows where to put it, it rolls downhill to the CISO. That happened with cloud security a decade ago. It's happening again with AI governance. The logic makes sense on the surface: CISOs live at the intersection of technology and risk, they know frameworks, and they already manage third-party risk and data protection. The problem is that AI governance is not a security problem wrapped in governance language. It is a business problem that includes security.
Think about what AI governance actually requires an organization to decide. Which use cases are acceptable and which aren't? How much risk will the organization tolerate in customer-facing AI versus internal tooling? How procurement evaluates AI vendors against criteria that don't exist yet in standard third-party risk questionnaires. Who takes the hit when a model produces biased outputs that trigger regulatory action? How training data gets sourced, retained, and audited against legal requirements that are still being written, including the EU AI Act's high-risk system obligations kicking in this August.
No CISO has unilateral authority over all those decisions. Most don't even have a seat at the table when procurement selects AI tools or when a business unit defines its own use cases. But when something breaks, the accountability chain points right at them.
The numbers back this up
Splunk's 2026 CISO Report surveyed 650 global security leaders and found that nearly all of them have absorbed AI governance responsibilities. 79% say their roles have expanded past what their mandates and resources can support. Meanwhile, only 6% of organizations running AI agents have updated governance frameworks to match what those agents actually do. Shadow AI was responsible for 20% of data breaches in 2025. And 71% of CISOs say AI has access to core business systems, but only 16% are governing that access well.
So the person holding the title doesn't have the authority, and the people who do have it aren't accountable for the outcomes. That's where the model breaks down.
What actually works
The organizations succeeding at AI governance aren't assigning a single owner. They're building governance operating models that spread the work across functions. In practice, it comes down to three things:
A governance charter that actually means something.
Not an acceptable use policy stapled to the employee handbook. A board-ratified charter that spells out decision rights, escalation paths, and risk tolerance thresholds across business functions. It says who approves high-risk AI deployments, who owns ongoing monitoring, and what triggers escalation to the executive team or the board.
A steering committee that reflects the actual risk surface.
AI governance committees need people from legal, compliance, data and analytics, procurement, HR, and the business lines deploying AI. Not just security. The CISO should be at the table, absolutely, but as one voice among several, not the sole owner.
The committees that work use RACI matrices to distribute ownership. Legal owns regulatory interpretation. Procurement owns vendor risk. Business units own use case justification. The CISO owns security and technical risk controls.
A regular operating cadence that keeps governance from becoming shelfware.
The organizations making this work run standing monthly reviews, quarterly recalibrations of risk tolerance, and have clear triggers for ad hoc escalation when something changes: a new regulation, a vendor shift, a shadow AI incident.
AI governance engagements
We build these structures in our AI governance engagements at Echelon Risk + Cyber: charters, committee designs, RACI matrices, and operating cadences. All tailored to the organization's size, industry, and risk profile. I've watched the difference between organizations that treat governance as an org design problem and those that just give the CISO another hat. It's night and day.
The industry's default answer to "who owns AI governance?" is wrong. Not because the CISO is the wrong person, but because making it one person's job is the wrong model.
AI risk spans too many functions for any single role to own and treating governance as a title assignment rather than an org design problem creates gaps.
If your AI governance strategy starts and ends with your CISO, you don't have governance. You have a bottleneck with a title. Does that sound like your organization? Echelon Risk + Cyber's AI Governance services are a good place to start.