· AI

Your AI Policy Is Not the Problem. Your Provisioning Queue Is.

Employees are not ignoring your AI policy out of carelessness. They are ignoring it because the unapproved option is faster than the approved one.

Formal AI policies nearly doubled this year, and employees kept ignoring them at almost exactly the same rate as before. ISACA's 2026 AI Pulse Poll, a global survey of more than 3,400 digital trust professionals, found that 38 percent of organizations now have a formal, comprehensive AI policy, up from 28 percent in 2025. That is real progress on paper. It has not moved the number that actually matters. PagerDuty's 2026 Shadow AI Survey, fielded by Wakefield Research across 1,250 office professionals at companies with at least $500 million in revenue, found that 66 percent of workers who use AI at work do so despite believing their employer's own policy does not permit it. That share rises to 72 percent at companies with 1,500 or more employees, the ones most likely to have a written policy in the first place.

My take: the policy document was never the real bottleneck, and writing a better one will not close this gap. The bottleneck is how long it takes an employee to get a sanctioned AI tool once they decide they want one. Fix that, and the policy starts to matter. Leave it alone, and next year's survey will show the same split no matter how carefully the policy gets worded.

The case for the document itself

That case deserves a fair hearing, because it is not a weak one. A written AI policy is the private sector's own answer to the kind of government mandate I would rather avoid: a company that sets its own rules on data handling, model access and human review gets ahead of a regulator who would otherwise write those rules for it, more slowly and with less understanding of the actual workflow. A policy also does legal work a sanctioned tool alone cannot. It is the document outside counsel reaches for first after an incident, and it sets the standard of care a court or an insurer will look for. ISACA's own trend line, that jump from 28 to 38 percent, is genuine movement toward that kind of institutional seriousness. A company with no written policy at all is worse off than one whose policy gets followed imperfectly.

But the gap is not closing, it is holding steady

The newer data says the thing a policy is supposed to fix, unauthorized use, is not actually tracking the thing meant to fix it. BlackFog's January 2026 survey of 2,000 employees at companies with more than 500 workers found that 51 percent had wired an AI tool into some other business system or app on their own, with nobody in IT ever signing off. That figure should worry a security team more than any raw adoption number, because it is not someone pasting a paragraph into a chat window. It is a live integration sitting inside company systems, exactly the kind of access a written policy cannot see, let alone govern after the fact. And 63 percent of the same respondents said unapproved AI use is simply acceptable when the company has not given them a sanctioned option to use instead.

Merge's 2026 State of AI Governance survey of 1,500 full-time US employees puts a number on the actual fix. Seventy-three percent say their company has no formal, enforced policy on AI access, and 57 percent say they would use AI more at work, through approved channels, if IT simply moved faster on provisioning it. That is a majority of a workforce telling you directly what closes the gap, and it is not stronger language in a PDF.

Run the comparison your employees are actually running

Think about the choice from an employee's side. A personal ChatGPT or Claude account costs nothing and takes under a minute to open. A sanctioned enterprise account, at most companies, means a procurement ticket, a security review, and a wait measured in weeks. When the compliant path is slower than the noncompliant one by that much, you are not fighting a discipline problem. You are fighting a queue, and a queue loses to free and instant almost every time, policy or no policy.

The fix costs real money, but less than most companies assume: a preapproved vendor list with single sign-on, a same-day provisioning path with logging and basic data loss prevention switched on, and a policy document that follows that rollout instead of sitting fifteen months ahead of it governing nothing. That ordering is most of what an AI governance engagement should actually produce: find out what your people are already reaching for, measured against numbers like BlackFog's and Merge's rather than guessed at, then build the fast, sanctioned lane before anyone finishes wordsmithing the policy that governs it. That is the sequence we run at Mojo AI Services. If your policy was written before anyone checked what employees were actually doing with AI, that is a half-day conversation worth having.

Sources

References used in this article. Links also appear alongside the relevant claims.

Let's talk

Let's make it happen.

You don't need a polished brief. A couple of lines about where you want your company to go is plenty, and we'll come back with what we'd tackle first.

We scope the work around your goals and timeline before quoting anything, so you know exactly what you're getting.

LocationBoca Raton, Florida
CoverageSouth Florida + remote nationwide
Status Now accepting clients