Rethinking the CISO role for an AI-saturated enterprise

When everyone owns AI, who owns the risk and the outcome?​

AI leadership is already distributed across the tech C-suite. CISOs, chief technology officers, chief information officers, and chief data and analytics officers all bring distinct priorities to AI strategy, delivery, performance, and accountability (figure 1). Add vendors and autonomous AI agents to the mix, and the lines between who makes a decision and who is responsible for its outcomes become less clear. 

From security gatekeeper to enterprise risk orchestrator

The growing prevalence of the CISO role suggests that organizations already see security and resilience becoming more important. Forty-nine percent of organizations in our survey report having a CISO role in 2026, up from 31% in 2023. Many CISOs are being measured on outcomes that extend beyond cybersecurity, including integration of security into AI initiatives, organizational security culture and workforce awareness, and business value enabled through risk reduction. AI could push that evolution further.

As AI agents become more autonomous and always on, users’ identity and access management could become core elements of the control plane for managing risk.⁵ The model is shifting from simply “who can access what” to “who (or what) can access what, at what time, and on whose behalf.” CISOs could lead their organizations’ effort to create transparency across the entire system, from managing identification and access processes for both AI agents and humans seeking access to tech platforms and solutions, to developing and owning the orchestration and control planes that manage systems of authority, control, and governance.

Monitoring will likely also need to adapt. Organizations need to be able to detect anomalous tool use, privilege escalation, unusual data access, goal deviation, unexpected interactions between connected systems, and changes in AI agent behavior. Controls should generate auditable logs and support rapid intervention when an AI agent crosses a defined risk threshold. Organizations should also establish in advance which actions require human approval, which can be automated, and which events trigger a pause, escalation, or investigation.

Detection and response also need to happen in near real-time, as AI operations—and the threats targeting them—move faster. As one CISO at a major American biopharmaceutical company says in an interview, “Back in the day, if a vulnerability got introduced, you had at least weeks. These days it’s getting down to hours.”

Given the role security needs to play in AI delivery, more emphasis likely needs to be placed on building in security from the start, as the AI solution is being designed and developed. Secure-by-design principles can work to embed real-time risk monitoring, audit trails, and explainability into the product life cycle—a task that could require more strategic collaboration between the product’s chief architect and the organization’s CISO, according to findings from Deloitte’s Global Future of Cyber Study.⁶

“Compromises of systems and applications and environments will happen,” the CISO at a healthcare company told us in an interview. “If you try to prevent every compromise, it’s going to be very hard to get business done in the company. The goal is you build your program so that if a system’s compromised, you see it as quickly as possible, you immediately contain it, and you eradicate it from the environment before it has a bigger impact on the company.”

The CISO’s role, then, is less about preventing or owning every risk AI might introduce and more about helping ensure that the enterprise can see the risks across functions, contain them, and intervene when necessary.

How CISOs can make shared ownership work

Agentic AI can turn fragmented accountability into real-time enterprise risk. Decision rights, escalation paths, ownership of autonomous-agent outcomes, and vendor risk all need remapping as AI distributes authority faster than accountability can be managed.

But cross-functional responsibility still requires clear accountability for outcomes. As risk leaders, CISOs can help define organizational risk thresholds, clarify ownership, and establish the right controls with shared responsibility across functions. Multiple functions might have a role in managing an AI use case, but one accountable owner can provide clarity about who is ultimately responsible for the outcome.

CISOs have an opportunity to develop and drive an AI-related risk management model across functions. These four moves can help clarify the CISO’s role in making that model work.

1. Assign clear ownership and decision rights. Each significant AI use case should have a clearly identified business owner, technical owner, and risk oversight owner. A decision-rights map should specify who can approve deployment, who can pause or restrict the system, who should be consulted, and who should be informed. Organizations should also establish in advance which actions require human approval, which actions can be automated, and which events should automatically trigger a pause, escalation, or investigation. The CISO can help define the security controls and escalation requirements around those decisions without becoming the default owner of the business outcome.

2. Bring cross-functional partners into AI decisions earlier. Security, architecture, data, privacy, compliance, procurement, finance, and business stakeholders should participate early in AI vendor and use-case decisions. Early collaboration can reduce downstream redesign, approval delays, and unmanaged implementation risk.

This is particularly important for third-party risk. Shifting tech-vendor pricing models and data-access terms are becoming sources of dependency risk. Vendor contracts can introduce risks related to data rights and intellectual property, while changing pricing models (such as imposing API fees, data-access charges, or token-based pricing) and data access terms can create new dependencies. As Hanan Szwarcbord, chief security officer at Micron Technology, a global provider of memory and storage solutions, says, “One approach can be making the tool selection a collaborative process across the organization, so it’s not just IT imposing one.”⁷

This convening role often falls naturally to the CISO, who already sits at the intersection of security, compliance, and vendor risk, and is likely well positioned to bring the right functions and vendors into the conversation.

3. Translate risk appetite into operating guardrails. According to a group chief information security officer in a major Japanese financial services group, “In many organizations, no one has really clearly defined what the cyber risk appetite should be, and how you’re going to calculate it, how you’re going to track it, and how you’re going to present it across the board.”⁸

As AI systems take on more autonomous actions, organizations need to translate risk appetite into concrete operating decisions. The CISO may not own those business decisions, but the function can help make them enforceable through controls, monitoring, and escalation mechanisms.

4. Test whether governance works at the speed of AI. Determine whether AI governance can function as an operating capability. Use scenario planning to test governance resilience. Regularly test high-impact scenarios such as vendor outages, compromised models, prompt-injection attacks, or unauthorized AI agent actions, and track governance effectiveness through measures such as ownership coverage, limiting system access to only what’s necessary, auditability, and incident response times. Think through owners, outcomes, controls, and operating models across different scenarios. This kind of scenario testing draws on capabilities many CISOs already run for security operations and incident response, making the security function a natural home for stress-testing AI governance.

According to the CISO at one healthcare company, “An annual strategy and planning exercise isn’t effective anymore. You have to consistently be revisiting priorities and understanding whether changes in the world around us have caused us to go back and pivot in our strategy.”⁹

AI governance effectiveness can also be measured through operational metrics. Useful measures could include the percentage of AI agents inventoried and assigned an owner, the percentage of AI agents with access limited to only the systems and data they need, the time required to revoke AI agent access, the percentage of high-impact decisions supported by audit trails, the number of unresolved policy exceptions, and the average time to detect and contain AI-related incidents..

Governing risk at machine speed

Organizations’ tech resilience can now depend on enabling security and managing risk at machine speed. But AI makes technology risk simultaneously a cyber, compliance, vendor, and operational responsibility, potentially leaving accountability gaps where those domains stop and start.

Many organizations will likely need to move from function-by-function ownership of AI risk management—cyber owns cyber, compliance owns compliance—to a single cross-domain risk management process tied to business outcomes. CISOs don’t need to own the governance of every possible risk AI could create, but they’re well positioned to connect security controls, monitoring, escalation, and resilience across functions so organizational leaders can see where risk exposure exists, who owns the outcome of an AI-related risk event, and when intervention is required. In an AI-saturated enterprise, the ability to make that model work at speed might become a defining part of the CISO role.