AI governance needs to become part of the CISO’s GRC program

COMMENTARY: For decades, cybersecurity leaders have been asked: What are we doing to manage risk?

The answer traditionally focused around vulnerability management, identity controls, security monitoring, incident response, compliance frameworks, and periodic risk assessments.

AI has changed that equation.

[SC Media Perspectives columns are written by a trusted community of SC Media cybersecurity subject matter experts. Read more Perspectives here.]

Organizations now deploy AI across nearly every part of the enterprise. Employees use generative AI to analyze information, developers integrate AI into applications, and security teams use AI for detection and response. 

Organizations have also deployed autonomous agents capable of accessing systems and taking actions.

For the CISO, AI no longer represents a technology issue, it’s a cybersecurity risk issue.

The battlefield has changed

A CISO may have a mature governance, risk, and compliance (GRC) program covering cloud infrastructure, endpoints, privileged access, third-party risk, and regulatory requirements. Then the organization introduces dozens of AI systems. Suddenly, the battlefield has changed.

An employee uploads confidential information into an AI service. A healthcare organization deploys an AI application that processes sensitive patient information. An AI agent receives permission to access internal systems.

None of these events necessarily look like a traditional cybersecurity incident, but each changes the organization’s risk profile.

If the CISO’s GRC program does not account for those changes, there's effectively a new battlefield that nobody monitors.

A policy Is not a control

Many organizations have responded to AI by creating acceptable-use policies. That's a starting point, but not a complete governance program.

Consider an organization that prohibits employees from entering confidential information into public AI systems.

How does it know whether that rule actually gets followed?

What happens when an employee connects an AI application to a corporate data repository?

Who determines whether that application has appropriate security controls?

What happens when an AI vendor changes its model or data-processing practices?

Who reassesses the risk?

These are GRC questions.

The CISO doesn’t need to own every AI decision, but the CISO should ensure the organization has a repeatable process for identifying, assessing, and managing the security risks those decisions create.

AI needs a risk assessment

GRC requires our teams to understand risk before accepting it.

AI should be no different.

Before deploying an AI system, organizations should ask:

Consider two AI systems. One analyzes publicly-available marketing information and drafts content. The other accesses customer records and can modify information in a database.

Calling both simply “AI” hides an enormous difference in risk.

The second system requires significantly greater scrutiny because its potential impacts are substantially greater. A mature GRC program lets organizations categorize AI systems according to risk and apply controls proportionally.

The agentic AI problem

The risk becomes greater as organizations move toward agentic AI.

A traditional AI application might deliver a recommendation. An AI agent can potentially execute it.

Imagine a security agent that detects a compromised account and automatically disables it. That's often a valuable defensive capability. But what happens if that agent incorrectly identifies a legitimate employee as compromised? What happens if an attacker manipulates the agent’s inputs? What happens if the agent has permission to disable accounts, but also access to systems it does not need?

Here's where least privilege becomes critical. An agent should have only the permissions required to serve its purpose and someone or something needs to continuously evaluate those permissions.

From annual assessment to continuous assurance

Traditional GRC programs often rely on periodic assessments. AI environments can change much faster. The risk assessment that was accurate three months ago may no longer describe today’s environment. This creates an opportunity to modernize GRC.

AI can help organizations continuously analyze evidence, identify changes and surface potential control failures, but while AI can automate risk management; it cannot eliminate the need for it. The CISO still needs to establish the framework, define acceptable risk, and make judgments when the situation warrants.

The CISO’s GRC program should offer visibility into the organization’s AI environment, establish minimum security requirements, identify material risks and ensure accountability when AI systems make consequential decisions.

We don't want to prevent the business from using AI. We want to make AI adoption defensible.

AI has become part of the enterprise attack surface. Therefore, it must become part of the enterprise risk program. CISOs should no longer ask whether AI governance belongs in GRC: They need to ask whether the GRC program has evolved quickly enough to govern the battlefield AI has already created.

Michael Petrov, chief executive officer, Digital Edge Ventures

SC Media Perspectives columns are written by a trusted community of SC Media cybersecurity subject matter experts. Each contribution has a goal of bringing a unique voice to important cybersecurity topics. Content strives to be of the highest quality, objective and non-commercial.