AI Strategy
99.7% Have AI Data Access Policies. Only 36% Can Prove Who Authorized an Agent's Action.
Most organizations have AI data access policies, but few can prove which person authorized an agent's sensitive action.

Table of contents
Everyone is talking about how much AI can do.
Almost nobody is talking about whether anyone can prove who said it was allowed to do it.
Delinea just gave us the clearest answer yet. And the numbers are worse than the headline suggests.
The report that should stop every board
Delinea surveyed 2,254 IT and security leaders and 2,250 non-IT employees across eight countries. The findings, published September 29, 2026:
- 99.7% of organizations have a formal policy governing what data AI tools and agents can access.
- 87% had an AI tool or agent access sensitive data beyond its intended scope in the past year.
- Only 36% of IT leaders can always trace a sensitive AI access event back to a named human authorizer.
- 76% of employees have bypassed the required approval process for using AI at some point.
- 60% have felt pressured to use AI on sensitive data even when they were unsure it was permitted.
- Only 51% say access is checked against those policies in real time.
The policy exists. The enforcement does not.
The same gap, measured two other ways
Kiteworks 2026 Data Security and Compliance Risk report found that 80% of organizations experienced at least one security or AI-related incident in the past year. 63% faced a compliance consequence, from audit findings to regulatory investigations. Their AI Governance Maturity Score averaged 35 out of 100.
Gartner's forecast is blunter: through 2028, more than 50% of AI initiatives will halt because of unresolved agentic identity challenges.
Three independent studies. One conclusion.
The problem worth naming precisely
A finance analyst recently found a $14,000 refund that broke policy. The employee who had the authority to approve had no memory of approving it. He was not lying. An AI agent had processed the refund under his login credentials, because that was how the workflow had been set up months earlier when he built it.
The approval was real. The human was not.
The mistake most organizations make is calling this a security problem. It is not. It is misattributed accountability. A human credential can authenticate the action. It cannot establish that the human made the decision. Until those two things are logged separately, you end up with an audit trail that is technically complete and evidentially misleading.
That is worse than having no record at all. A missing record looks like a gap. A record that points at the wrong person looks like evidence.
Why this keeps happening
1. The policy and the enforcement live in different places.
A policy document says an agent can read customer records but not write to them. The policy lives in a PDF. The agent lives in a production environment with whatever credentials it inherited from the engineer who built the workflow. Nothing connects the two.
Delinea's CEO put it simply:
Written policy is only as good as your ability to enforce it at the moment an AI agent acts.
2. Accountability points at the login, not the decision.
When an agent acts under an employee's credentials, the audit trail says a person did it. That person may have set up the workflow months earlier and never touched it since. They are accountable on paper for decisions they never made and often never see.
Only 36% of organizations can trace a sensitive access event back to a named human authorizer. The rest have records that look complete and point at the wrong person.
3. Employees route around governance because governance is slower than the deadline.
60% of employees have felt pressured to use AI on sensitive data even when they were unsure it was permitted. 76% have bypassed the approval process at some point.
This is not a discipline problem. It is a design problem. When the approval process takes days and the deadline is today, the rational move is to skip the process. The governance gap is not that people break the rules. It is that the rules were never designed to survive contact with a real deadline.
The role most organizations never assigned
Every AI rollout has a procurement owner. A technical implementation owner. A security owner. A budget owner.
Almost none have an authorization owner. Someone whose job is to confirm, at the moment the agent acts, that a named human approved that specific access. Not the login. The decision. Recorded at the point of action.
Without that role, the audit trail is not a record of accountability. It is a record of access.
Access and authorization sound like the same thing. They are not.
Access is who could have acted. Authorization is who said this action was allowed. One is a fact about permissions. The other is a decision. Only one of them can be held to account.
The AI Value Equation, extended for enforcement
The five questions from earlier issues still apply. There is a further one that has to be answered before an agent runs, not after it fails:
Who authorized this action, at the moment it happened, and can you prove it?
Not "who built the workflow." Not "whose login did it use." A specific, named person, recorded at the point of action, with the authority to approve that specific access.
If the honest answer inside your organization is "we can probably figure it out after the fact," you do not yet have enforcement.
You have a policy document and a hope.
Decision of the Week
Before your next AI deployment, ask one question in the room:
If an AI agent accessed sensitive data tomorrow, how long would it take you to name the person who authorized it?
Separately, Delinea found that 55% of organizations take a full day or longer to detect a scope violation. That detection lag is a different measure from whether an organization can trace an access event to its human authorizer.
The Strategic Edge: AI governance is not about writing policies. It is about proving who authorized an action when something goes wrong. Access tells you who could have acted. Authorization tells you who said it was allowed. Only one of those can be held to account.
A note on sourcing: Delinea's AI Enforcement Gap report, published September 29, 2026, surveyed 2,254 IT and security leaders and 2,250 non-IT employees across eight countries. Kiteworks' 2026 Data Security and Compliance Risk report surveyed security and compliance leaders globally. Gartner's AI agent reality check was published October 2, 2026.
Thanks for reading Issue #8 of Strategic Edge.
Every week, I will break down a real business event, not to report the news, but to uncover the strategic decisions hiding beneath it.
Because in the AI era, technology is becoming abundant. Sound judgment is not.