The “responsible governance” framing isn’t neutral. The cost of compliance is a skill gap nobody is measuring.
Last Friday I was on a monthly call I host — sales ops leaders, marketing directors, RevOps people, content strategists — most of them in or adjacent to marketing operations, across companies ranging from twenty employees to enterprise scale. One of them, a senior person at a major tech company, said something quietly devastating: “I feel ignorant on AI.”
He isn’t. He’s smart, experienced, technically competent. He just can’t use AI in any meaningful way at work. His IT team has restricted access to a single sanctioned AI tool — an internal model walled off from the open web, locked to the company knowledge base, designed to be safe in the way a fenced playground is safe for a five-year-old. He can’t get it to do any of the work he actually needs to do.
Meanwhile, on the same call, a sales leader at a less-restricted company demoed a working SDR forecasting app he’d built in four hours. It replaces a Google-Sheets-and-manual-process workflow that his peers at much larger companies are still running. Same kind of role, same kind of person. Completely different worlds.
And here’s what’s underneath that: the sandbox isn’t neutral. Someone who’s never used AI for work has no impression of it. Someone whose only experience is a hobbled sandbox tool has a strong one. And the impression is that AI doesn’t really work. Their employer has spent months teaching them not to trust the most consequential tool of their career.
This is the conversation that isn’t being had honestly: the “responsible AI governance” narrative — currently the default take in every consultancy white paper, vendor pitch, and HBR explainer — assumes that sanctioned enterprise tools work just as well as external AI. They don’t. The gap between someone who can experiment with frontier tools and someone who can only use a corporate sandbox is widening every week. And it isn’t the kind of gap your IT team thinks they’re managing. They believe they’re managing a privacy gap. What they’re actually creating is a skill gap.
Most people in this position read their constraint as governance maturity — my company is being responsible. I want to push back on that. Your AI lockdown isn’t risk management. It’s a preview of how this company handles change. And you’ve been reading it wrong.
Someone Is Selling You This Position
The reason “balanced governance” is so loud right now isn’t that it’s been carefully validated. It’s that Microsoft has spent the last year explicitly making enterprise AI governance their selling point. The Copilot Responsibility Framework, announced in March 2026, positions governance itself as the differentiator that makes Copilot the enterprise-safe choice over OpenAI’s and Anthropic’s offerings. Salesforce, SAP, and Oracle have followed with their own governance frameworks in response.
In other words: the discussion about whether your IT team should restrict you to sanctioned tools isn’t a neutral conversation about security. It’s a marketing fight. The companies fighting it are the enterprise software vendors who benefit from making “sanctioned tooling” the default frame for responsible AI use. When you hear someone say “organizations need clear AI policies, sanctioned tooling, and employee training,” you’re hearing a position that’s also being sold to you.
That doesn’t make the position wrong. There are real privacy risks, real compliance concerns, real reasons to have controls. It does mean that the person whose IT team has chosen the most restrictive interpretation of this advice has been handed someone else’s sales pitch as if it were a universal truth. The default take is one answer to the AI governance question. It’s being marketed as the answer. Those aren’t the same thing.
This is the bookend to a post I wrote nine months ago about Shadow AI, the failure mode when leadership doesn’t lead on AI policy and employees use ChatGPT in secret. Same governance question, opposite failure. There, leadership didn’t lead. Here, leadership led so completely that it took the tools away.
The Gap Is Already Here. You’re On the Wrong Side.
Notably absent from the “balanced governance” conversation: data on what happens to people who comply with it.
The MIT NANDA report The GenAI Divide: State of AI in Business 2025 found that only 40% of surveyed organizations have purchased an LLM subscription, while 90% of employees regularly use personal AI-powered tools for work tasks. MIT describes this as a “shadow AI economy” that is more effective than the official one. Only 5% of corporations report any economic benefit from their official AI implementations. The report’s recommendation: organizations should learn from shadow usage and analyze which personal tools deliver value before procuring enterprise alternatives. That’s a polite academic way of saying the employees are right and the policy is wrong.
Anthropic’s own economists named the same pattern in March: the AI skills gap is forming, and power users are pulling ahead. A separate NBER working paper from February (Gálvez & Lombardi) measured the productivity effect quantitatively. Without AI, a high-education worker outperformed a low-education worker by 0.548 standard deviations. With AI access, the gap fell to 0.139, closing about three-quarters of the baseline. The mirror reading is the one your IT team isn’t doing: removing access reopens skill gaps that AI was closing.
If you’re the one being locked out, this is the part of the conversation nobody is having with you. The cost of sanctioned-tool-only governance isn’t a marginal productivity hit. It’s a skill gap widening every week between you and your less-constrained peers.
I wrote three weeks ago about the learning penalty, the cost paid by junior people who never have to do the messy work that builds judgment because AI does it for them. This is the same skill gap from the opposite direction. Junior people using AI too much fail to build judgment. Senior people blocked from AI entirely fail to develop fluency with the tool reshaping their profession. Same erosion. Opposite cause.
Read the Lock
Here’s where it gets useful. If your company’s AI policy locks you to a sanctioned tool that can’t do real work, that policy is telling you three things. Most people are only hearing the first one.
One: my IT team is risk-averse. True. Also not particularly informative. Every IT team is risk-averse.
Two: my IT team picked the most restrictive interpretation of a debate where reasonable people disagree. Also true. More informative. This is a choice, not an inevitability. Other companies, including very large and very regulated ones, are choosing differently.
Three: my company’s leadership has either not asked, or has asked and accepted, that their people can’t develop AI fluency through the only mechanism that develops it: access to actual frontier tools used on real problems. This is the part nobody’s saying out loud. It’s information about your company, not about your IT team. It’s information about how your leadership picks between feeling safe right now and being able to keep up later. And if you’re sitting inside that decision, you should be reading it.
This isn’t a story about whether your company is “behind on AI.” Plenty of companies are behind on AI. This is a story about what happens to you — to your skill development, to your market value, to your judgment — when your employer has effectively decided that your professional growth matters less than picking the safest side of a marketing fight.
That’s the part you’ve been misreading. The lock isn’t responsible governance. It’s a preview of how this company will compete over the next three years. And whether the people who work here will still be able to compete with the people who don’t.
What You Can Do About It
You can read this signal and respond to it. Three moves, easiest to hardest.
1. Find the slim space within sanctioned tooling. Try this first.
Someone I work with — at a company with similar restrictions to the one I described at the top — was dealing with a textbook operational problem. Sales reps in her company’s CRM had to follow a flowchart so dense for customer-object creation that they got it wrong every week. Her team spent hours cleaning up the records. The flowchart was the bottleneck. The restricted AI access wasn’t. So she used her sanctioned AI tool — the one supposedly limited to internal use — to build a step-by-step wizard that walked reps through the same decision tree. A “choose your own adventure” interface. Deployed it. Hours of data cleanup work disappeared every week. No policy violated. No expanded access required.
The reframe works when the actual constraint is “I need any AI to help me solve a specific operational problem” rather than “I need access to GPT-4.” If you can frame your problem as a design problem rather than a capability problem, the sanctioned tool may have more reach than you’ve given it credit for. Try this before you give up.
2. Build the case up. But reframe the conversation. This is the move most people are told to make and almost none of them do well. The reason: they argue from capability. “We need ChatGPT Enterprise because Copilot can’t do X.” That argument loses, because the people approving the policy aren’t evaluating capability. They’re evaluating risk.
Frame the case as skill-gap risk. The Anthropic data is real and citable. The NBER paper is real and citable. The fact that your peers at less-restricted companies are pulling ahead is a hiring and retention problem for your company. The people who can demonstrate fluency with frontier tools won’t be at your company, because they can’t develop that fluency at your company. The argument lands in a different conversation (talent, retention, who can hire and keep good people) than the usual one (security, compliance). Make the policy conversation about what the current policy is costing the business, not about which tool you want next.
If that conversation has been had and rejected, you have new information about your company. Move to step three.
3. Read the signal honestly. The most important version of this move is also the easiest to misread, so I want to say it plainly: I’m not telling you to go work for a startup. I’m not telling you that smaller companies are inherently better. I’m telling you that an unmovable AI lockdown is information about how your company handles change, and you should treat it as such.
If your company has decided that their people can’t develop fluency with the most consequential tool reshaping knowledge work, the question to ask is: what does this tell me about how this company will respond to the next thing? AI is the visible question right now. It won’t be the only one. The same leadership instinct that produced this lockdown — pick the safest option, worry about the future later — will show up again. Two years from now it’ll be agentic operations. Three years from now it’ll be something else. You’ll be having the same conversation with the same outcome.
This isn’t envy of people at less-constrained companies. It’s information about where the company you currently work for is going, gathered from the clearest signal you have. Treat it that way. You can decide to stay anyway. There are excellent reasons to. But make that decision with the information, not in spite of it.
The person from Friday’s call ended his comment by saying he was thinking about writing an article on AI’s impact on language and communication. He was going to do it from the outside, observing. Because from the inside, he couldn’t really see it.
That’s the cost. He’s a smart person whose primary access to the AI revolution reshaping his profession is the same access a stranger would have. And his company hasn’t noticed that this is a problem.
The lock is the message. Read it.
Sources and Further Reading
- “The AI skills gap is here, says AI company, and power users are pulling ahead” — TechCrunch, March 26, 2026
- Gálvez & Lombardi, “Does Generative AI Narrow Education-Based Productivity Gaps? Evidence from a Randomized Experiment” — NBER Working Paper #34851, February 2026
- MIT NANDA, The GenAI Divide: State of AI in Business 2025 — July 2025
- “Enterprise AI Governance in 2026: Why the Tools Employees Use Are Ahead of the Policies That Cover Them” — MarkTechPost, May 13, 2026
Related Reading from Pallas Advisory
- Your Employees Are Already Using AI (And They’re Not Telling You) — the bookend opposite. The failure mode when leadership doesn’t lead on AI policy.
- AI Productivity Has a Hidden Cost. It’s Called the Learning Penalty. — the same skill gap from the opposite direction. Juniors losing judgment from too much AI use; this post describes seniors losing fluency from too little.