AI governance tends to be treated as a specialized discipline. A central AI team might own the program. Data leaders could manage inventories and standards. Legal and compliance teams could interpret emerging requirements. Security had an important role, but it was often one participant among many.
And that model is becoming harder to maintain.
AI is moving deeper into everyday business operations. It’s appearing in third-party applications and internal workflows, and agents are beginning to take actions instead of simply producing content. The security implications now extend far beyond whether a model passed an assessment before deployment.
This doesn’t mean the CISO should suddenly own the entire AI governance program. But it does mean AI governance is becoming inseparable from the responsibilities security teams already have.
The question isn’t whether security belongs in AI governance; it’s how deeply the two functions need to connect.
AI Adoption Has Moved the Risk Conversation
The scale of AI adoption has changed what governance requires.
Nearly three-fourths (74%) of enterprises have reached departmental or scaled AI adoption. More than half are using AI across multiple functions or embedding it into business processes. At the same time, just 5% say coordination and accountability are clear across the AI lifecycle.
That disconnect is a huge concern for security teams.
AI can introduce new paths to sensitive information. It can change how applications interact with enterprise systems. Third-party AI services can bring models and data sources into workflows without the same visibility organizations have over internally developed technology.
Security teams already understand this type of problem. You can’t protect an environment you can’t see, and you can’t manage risk without knowing who owns an asset or what it connects to.
AI governance brings those familiar principles into a much more dynamic environment.
The AI Inventory Is Becoming Part of the Security Picture
Traditional security programs depend on visibility. Teams maintain inventories of systems and identities. They track vendors and vulnerabilities. They monitor access and understand dependencies.
AI introduces another layer.
Organizations need to understand the AI systems operating across the enterprise. That includes internally developed AI as well as third-party and embedded capabilities that include models and autonomous agents.
For security leaders, that should sound familiar. Shadow AI is the new generation of Shadow IT.
The goal is not simply to create a longer inventory. Security teams need enough context to understand what an AI system does and what it can reach. They also need to understand its ownership and risk profile.
Third-Party AI Makes the Boundary Even Harder to See
Security teams have spent years expanding their focus beyond the traditional enterprise perimeter. AI accelerates that shift.
An organization may not build the model behind an AI capability. It may not operate the infrastructure. The capability could arrive through software the business already uses.
That doesn’t remove the organization’s responsibility for how the technology is deployed.
Third-party AI governance requires teams to understand the source of an AI capability and the data it uses. They also need visibility into vendor changes and security evidence. Ongoing oversight matters because the capability can change after the original review.
More than a third of organizations (39%) say limited visibility into third-party AI, vendors, or data sources is a factor into delaying AI initiatives. More than half (52%) pointed to concerns around data quality, access, privacy, or security.
These aren’t isolated governance issues. They sit directly beside third-party risk management and information security.
Agentic AI Raises the Stakes Again
Generative AI made governance more urgent. Agentic AI makes it more operational.
An AI agent can interact with systems and tools. It can use data and initiate actions. That changes the security conversation from what an AI system produces to what it’s permitted to do.
Organizations are moving quickly in this direction. For security teams, agentic AI brings familiar questions into unfamiliar territory:
- What resources can the agent access?
- What permissions does it have?
- What data can it consume?
- What happens when its behavior changes?
- Who responds when a control fails?
Those questions can’t be answered through an AI policy alone.
Approval Is No Longer the Finish Line
Many AI governance programs were designed around intake and approval. A use case is submitted, teams assess the risk, and controls are documented. Someone approves deployment.
That’s static governance. AI systems require dynamic oversight and management.
Models change. Data access changes. Tools change. Agent behavior will change. The system operating six months after approval won’t look exactly like the system that was reviewed.
That’s why security teams have an increasingly important role after deployment.
Governance needs to extend into ongoing operations. Organizations need ways to identify meaningful changes and determine whether controls remain effective. They need clear ownership when something requires investigation or remediation.
This is where AI governance begins to look much more like the continuous risk practices security teams already know.
Security Can’t Do This Alone
Making AI governance part of the security agenda doesn’t mean turning it into another security silo. The opposite is what’s actually needed.
AI governance crosses organizational boundaries by design. Successful initiatives require alignment across data, security, risk, IT, privacy, and AI stakeholders.
Security brings an essential perspective to that group. It understands controls and incident response. It understands identity and third-party exposure. It knows that a policy has limited value when nobody can determine whether the control behind it is working.
AI governance brings the context security needs to apply those disciplines appropriately. It connects technical signals to the purpose of an AI system and its owners. It also connects them to policy requirements and business risk.
That connection is critical.
AI Governance Must Be Part of the Security Operating Model
Security teams don’t need to become AI scientists. They don’t need to take ownership away from data teams or AI leaders. But they do need a seat at the table that extends beyond the initial risk review.
AI is becoming another fundamental layer of enterprise technology. As that happens, AI risk becomes intertwined with information security and identity. It also connects to third-party risk and broader enterprise resilience.
The security organizations that adapt will treat AI governance as a shared operating responsibility rather than a separate compliance exercise.
That’s the larger shift currently underway. AI governance is moving closer to where AI operates — and security is already there.
