Enterprise AI Governance Is Missing Its Third Layer

Joseph Ours leads the AI Strategy Practice at Centric Consulting.

​Gartner researchers project that 40% of enterprise applications will embed task-specific AI agents by the end of this year, but many organizations racing toward that number have governance frameworks that weren't designed for it.​

Those frameworks were built to answer two questions: Does the technology work, and has it been approved internally? Although both matter, a March 2026 federal court ruling established that a third question now carries legal weight: Does the platform your agent is operating on permit it to be there?

For most enterprises, the question hasn't been asked and, worse, it can't be answered.

Most organizations have invested in model risk management, human-in-the-loop checkpoints, role-based access controls and deployment review boards. However, according to McKinsey's 2026 AI Trust Maturity Survey of approximately 500 organizations, agentic AI controls lag behind every other governance dimension, with just 30% reaching meaningful maturity in that area.​

Those frameworks were designed for AI systems operating within organizational boundaries, on internal data, against internal systems and under internal supervision. By design, agentic AI operates across those boundaries. Agents interact with supplier portals, partner platforms, SaaS tools and third-party data sources as part of their standard operations.

I’ve found that there are two governance layers that cover internal operations. The first, technical capability, tells you what the agent can do. The second, organizational governance, tells you what it's been approved to do. Neither address whether the external platform the agent is working against has authorized that activity.

That third layer, platform authorization, is what most governance programs haven't built. Gartner researchers forecast that more than "40% of agentic AI projects will be canceled by the end of 2027, due to escalating costs, unclear business value or inadequate risk controls." I've found that those "inadequate risk controls" often include platform authorization gaps.

Platform authorization governance is an operational function, not a legal one. It belongs in the deployment process alongside security reviews and data classification, rather than in a vendor onboarding checklist.

1. A Platform Interaction Inventory: Governance requires visibility into where agents operate. That means mapping every instance where an agent interacts with a third-party system, including unofficial deployments. Shadow agent usage is a significant exposure vector. A sales team running browser automation against LinkedIn without IT's knowledge surfaces in an inventory audit, not a security incident.

2. A Terms-Of-Service Review Cadence: For every platform an agent touches, governance teams need a documented review of current terms covering automated access, agent activity, bot restrictions and data extraction. Platforms are actively revising their terms in response to the evolving legal landscape, which means the review needs to be recurring, tied to a schedule and triggered any time a new agent deployment targets an existing platform. Treat it the way security teams treat vulnerability scanning: The environment changes, so the review has to keep pace with it.

3. Platform Authorization As A Deployment Gate: Before any agent deployment that interacts with an external system, verification that the target platform permits agent access is a hard requirement, the same as a security review. In practice, this means adding a platform authorization check to the agent development life cycle, sitting alongside the existing security and compliance gates that fire before production deployment. The check is documented; it has a named owner, and it doesn't get waived because the timeline is tight.

Platform authorization tends not to get built because nobody owns it. Technical teams think in terms of capability and access. Legal and compliance teams think in terms of contracts and regulations. Platform authorization sits between those functions—operational, contractual and continuously changing—and it falls through the cracks accordingly. The result is a deployment review process that asks every hard question about internal risk and almost none about external authorization. By the time the questions surface, the agent is already in production.

Organizations that have closed this gap assign platform authorization as a dedicated function within their AI governance program, with explicit ownership, a review cadence and integration into the deployment life cycle. The investment is process design and accountability, not headcount. Without a named owner, the inventory doesn't get built, the terms don't get reviewed and the deployment gate doesn't get enforced.

Gartner analysts identify "loss of control," or agents operating outside appropriate constraints, as "the top concern for 40% of Fortune 1000 companies by 2028." Platform authorization is one of the most concrete and addressable forms of that risk.​

The riskiest deployments are those that work exactly as designed, operating against platforms that didn't authorize them and accumulating exposure until it can't be ignored. Enterprise AI governance programs are well-constructed for the environment that existed two years ago. Building the third layer is how organizations keep their agent programs running through what comes next.

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?