Authored By Marc Shorb
When leaders sit down to write an AI policy, the goal is usually practical. Set some guardrails, reduce risk, and keep people from doing something that lands the company in trouble. The output is a document about what’s allowed and what isn’t.
But employees don’t just read the rules. They read into them. Every policy carries an implicit message about how leadership feels, and people absorb that message whether or not anyone meant to send it. A recent Founder Reports survey of 2,078 U.S. workers suggests that with AI specifically, the way you write the policy may be shaping how much your team trusts each other’s work, in ways HR almost certainly didn’t plan for.
The signal in the data
Here’s the finding that started this line of thinking. Workers at companies that allow AI with restrictions distrust AI-assisted work more than workers at companies that allow it freely. Not a little more. Among the restricted group, 52% say they trust a coworker’s AI-assisted output less. Among the unrestricted group, that figure is only 28%.
That’s a wide spread, and it runs opposite to what you might expect. You’d think clearer rules would make people more comfortable, not less. Instead, the more locked-down the policy, the less employees seem to trust the output it governs.
Restrictions function as a signal. When an organization feels it has to limit AI use, employees interpret that as an admission that the output carries risk. The policy was meant to manage behavior, but it also quietly answered a question nobody asked out loud: is this stuff trustworthy? The restriction itself says, “no.”
Policy as communication, not just rules
Step back from the numbers and the broader point comes into focus. HR tends to think of policy as governance. What’s permitted, what’s prohibited, what happens if someone crosses the line.
But employees experience policy as a statement of values. A restrictive AI policy says, between the lines, “we don’t fully trust this.” A permissive one says “we’re comfortable with this.” Nobody writes those sentences into the document, but people read them anyway, and then they act on them.
Your policy is not only regulating what people do with a tool. It’s also setting the emotional temperature around a tool your workforce is already using constantly. The same survey found that 89% of workers have used AI for their jobs and 60% use it daily or weekly. This is a daily reality you’re framing for people, and the frame you choose colors how they see one another’s work.
The catch: restrictions aren’t the enemy
None of this means restrictions are a mistake. Sometimes they’re exactly right.
Plenty of work involves sensitive data, regulated information, or stakes where an AI error carries real consequences. Limiting AI in those situations is just good judgment. And the caution isn’t irrational in general, either. The same survey found that 45% of workers have had to fix or redo a coworker’s work that relied too heavily on AI. People have a real basis for being careful.
So the problem isn’t restriction. The problem is unexplained restriction.
A limit that shows up with no context reads as a blanket verdict on AI. The same limit, with a reason attached, reads as a considered boundary. Consider the difference between these two:
- “AI use is restricted.”
- “We limit AI use on client financial documents because of confidentiality requirements.”
The first one signals distrust across the board. The second one signals careful thinking about one specific risk. The rule is similar. The message is completely different. What changes the signal isn’t whether you restrict, it’s whether you explain.
How to write policy with the signal in mind
If a policy is always communicating something, the move is to communicate on purpose instead of by accident. A few ways to do that:
- Explain the why, not just the what. A restriction with a reason sends a targeted message. A restriction with no reason sends a global one. The explanation is where the actual signal lives, so don’t leave it out.
- Match the policy to the real risk, not the general anxiety. Blanket restrictions communicate blanket distrust. Scoped restrictions communicate judgment. If only certain work is genuinely high-risk, restrict that work specifically, so you’re not broadcasting suspicion across everything your team produces.
- Don’t let silence become the policy. In the survey, 44% of workers said they have no clear AI policy or aren’t sure one exists. Silence isn’t neutral. People fill the gap with their own assumptions, and those assumptions are something you no longer control. No message is still a message, and usually a confusing one.
- Revisit the tone as adoption matures. A policy written in a moment of early caution can keep sending that cautious signal long after the organization has actually grown comfortable with the tools. The reality evolves. The document should too.
None of this is about loosening every rule. It’s about making sure the rules say what you mean.
Say what you actually mean
HR can’t control every inference an employee draws. But the alternative to controlling the message is sending one by accident.
An AI policy is always saying something about how the organization feels about AI. The only real choice is whether that something is deliberate or unintended. Right now, a lot of policies are quietly telling employees not to trust work the company itself permits, and no one decided to send that message. It just came along for the ride.
The best AI policies explain themselves, so the signal and the intent finally line up. Write the policy you mean, and make sure it reads the way you meant it.
Author Bio
Marc Shorb is the founder and editorial manager at Founder Reports, a business and entrepreneurial-focused publication. Founder Reports provides insight for business owners and leaders through original studies, in-depth reports, and interviews with industry leaders.
