"What is AI compliance?" Owners ask it when a customer's questionnaire arrives with a section on artificial intelligence, when an insurer or a bank adds a line about it, or when a board member says the company should be "compliant" and nobody can say with what. The word sounds like a certificate you buy or a rulebook someone else has written. For an owner it is simpler than that. This page gives it one meaning an owner can use, and a short list of what to decide and write down. It is general information for owners, not legal advice; which rules apply to your company is a question for your own lawyer.

I was COO and then President of a public company that closed more than forty acquisitions, and I ran the integration on most of them. In every one of them somebody eventually asked whether the company had done what it said it did, and the companies that answered well were not the ones with the most paperwork. They were the ones that could put a dated page on the table. I am a lawyer by training, which is why I am careful about where this page stops. Now, at 013 Labs, I write fixed-fee AI plans for CEOs of private-equity-backed and owner-led mid-market companies, and the compliance question is one of the first that comes up.

The short answer

AI compliance is doing what the rules and promises that already apply to your company require when you use AI, and being able to show it. Two halves, and both matter. The doing is mostly the ordinary obligations your company already has, applied to a new kind of tool: how you handle customers' and employees' information, what you promised in contracts, what you told your insurer and your bank, what the law in your sector and your location requires. The showing is a dated page, with names on it, that a person who was not in the room can follow.

There is no single AI rulebook that covers every company. What applies to you depends on what you do, where you do it, who you serve and what you have signed. That is why the first move is not to buy anything. It is to find out which rules and promises already touch the work where AI is used.

What AI compliance is not

It is not a certificate. A vendor may tell you its product is "compliant". That statement is about the vendor's product, under conditions the vendor chose. It does not tell you whether your use of it, with your customers' information and your staff, meets the promises you made.

It is not the same as an AI policy. The page on whether you need an AI policy is about what your staff are told to do. Compliance is whether the company, as a whole, does what it has promised to others. A policy is one of the things that helps you show it.

It is not a project with an end date. Contracts renew, questionnaires change and tools add features in updates. The page is kept alive or it is useless.

The three kinds of rules

An owner can sort the whole subject into three piles. Each has a different source and a different person who knows it.

The law that already applied. Rules about how personal information is collected and kept, about how employees are treated and hired, about how customers are told the truth, and any rules that are specific to your sector. Most were written before the current tools and still apply to them, because the law usually follows what you do, not what you do it with. Whether any newer rule aimed at AI reaches your company is a question for your lawyer. Nothing here summarises any particular law.

The promises you made. This is the pile most owners forget, and it is often the one that bites first. Your customer contracts and questionnaires, your insurance application, your loan agreement, your agreements with partners who share information with you, the terms you accepted when your company signed up for each tool. Each is a promise about information, quality or who does the work, and each can be broken by a well-meant use of a tool.

Your own rules. The data rule, the list of what a tool never does alone, and the written scope of whoever owns the question. These are the rules you control, and they are the easiest part to write. They are also the thing every outsider will ask for first, because they show that a person has decided.

Who asks for proof

The question reaches a company from five directions, and none of them is a technology question.

The customer. Larger customers send questionnaires to their suppliers, and artificial intelligence is now a section in many of them: whether you use it on their work, whether their information goes into it, who checks. The page on what an AI disclosure is covers how to answer in a true and short statement.

The insurer. Renewal forms and underwriters ask whether you use AI and how it is controlled. The page on whether your insurance will ask about AI covers what to expect.

The bank. A lender may add a line about it to a covenant or a review. The page on what to tell your bank about AI covers that conversation.

A buyer. In diligence a buyer asks whether the company knows what it uses, what it has promised and who is answerable. The page on whether AI makes a business worth more when you sell covers what a buyer looks for.

A regulator or an auditor. In regulated sectors this may arrive as a request for records, and your lawyer should be on the first call.

All five want the same thing: proof that a person decided and a person checks. One dated page answers all five.

The three decisions behind it

The same three decisions that sit under every AI plan sit under compliance. Only the questions change.

Who owns the question. One named person, with a half-page written scope: what they may approve alone, what comes to the owner, and when the page is reviewed. In a company this size it is often the finance chief, the operations lead or the person who already handles contracts and insurance, with the company's lawyer as the adviser and not the owner. The page on who owns AI in a company is about choosing that person.

What information stays out. One sentence the whole company can repeat. For most companies it names customers' information, anything covered by a confidentiality promise, employee records, and anything under a non-disclosure agreement, and says none of it goes into a tool the company has not approved, on an account it does not hold. Most broken promises are about information. The page on whether free tools are safe for business use covers the account and data side.

What the tool is never allowed to do alone. A short list, three items is a good number. Nothing is sent to a customer, signed or filed, and no decision about a person is made, until a named person has looked. The page on human in the loop is the long version of that check.

If you cannot write each decision in a few lines, you have found the real gap. That is a useful thing to learn before the questionnaire arrives.

The one page

A compliance record needs a single page, kept where the owner and the person who owns the question can find it. Six lines is enough.

What is in use. By team and task, including the free tools and the features a software vendor switched on in an update. This is the line most companies cannot fill in on the first try, and it is the one every outsider asks about first.

What applies. The rules and promises that touch those uses, listed in plain words: the customer contracts that mention information or quality, the insurance and loan terms, and the sector rules your lawyer has named. Do not write what you think the law says; write what your lawyer told you applies, and the date.

Who owns it. The name and the written scope.

The data rule and the never-build list. The sentence and the three items, as above.

The checks. Who looks at what before anything leaves the company, and where the record of that is kept.

The review date and the signature. The owner signs, the owner of the question keeps it, and whoever answers questionnaires attaches it.

Check yourself

Seven yes or no questions. Each no is something to decide before someone asks.

Mostly no? That is the normal starting point for an owner-led company. The AI readiness scorecard asks ten questions like these across the whole business and shows which of the three decisions is missing.

How often, how much, who helps

An owner's next questions are usually about effort, and the honest answers are short.

How often: at least once a year, and again whenever a major customer sends a new questionnaire, the insurance or the loan renews, or a new tool or kind of information arrives. A quick quarterly look at what is in use catches most of the drift.

How much work: a first version takes an afternoon of the owner's time and a few hours of the owner of the question, mostly spent finding out what is in use and which contracts say what. Most of the effort is looking, not writing.

Who helps: your lawyer for what applies, your insurance adviser for what the policy expects, your finance chief for what the loan says. Be wary of anyone who offers to make you compliant in a week with software.

When someone asks

The day a questionnaire or a request arrives, three sentences should be possible. Here is what we use, and who checks it. Here is the rule about information. Here is the page, dated, with a name on it.

If an answer on the form is not yet true, do not write it as if it were. Say what is true today, what is changing and by when. A true "not yet, here is the date" is worth more than a smooth yes that falls apart on the first follow-up. If you do not know whether a particular rule applies, say so and call your lawyer. Guessing in writing is worse than a gap.

Picture it going wrong

A made-up example, to show the shape. A sixty-five-person regional freight broker has been using free tools for a season. The dispatch team drafts customer emails with one, the office summarises carrier contracts with another, and a manager pastes rate sheets into a third to check them. Nothing is written down. Then the company's largest customer sends its yearly supplier questionnaire, with a new section on artificial intelligence: whether it is used on the customer's shipments, whether the customer's information goes into it, and who reviews the output. The owner cannot answer any of the three. The contract the company signed with that customer contains a promise about confidentiality that two of the three tools may already have strained.

Nothing in that story is a technology failure. No one owned the question, no one had written the data rule, and no one could say what was in use. With one page written, the freight broker answers the questionnaire in a day: here are the four tools, here is the owner, here is the data rule, here is what a person checks, here is the review date. Where a promise was strained, the owner says what changed and when.

Mistakes that repeat

Frequently asked questions

What is AI compliance?

Doing what the rules and promises that already apply to your company require when you use AI, and being able to show it with a dated page. It is mostly your existing obligations on information, contracts and honesty, applied to a new kind of tool. This is general information, not legal advice.

Do I need AI compliance for my small business?

You need the answer to the underlying question: do you know what AI is in use, what you have promised others about information and quality, and who checks. A very small company may need only a short page and a conversation with its lawyer. Whether any particular rule applies to you is for your own lawyer to say.

Is there a law I have to follow for AI?

It depends on where you operate, which sector you are in and who your customers are. Some rules aimed at AI exist or are coming in some places, and many older rules about information, employment and consumer honesty apply to AI use already. Ask your lawyer which apply to your company. Nothing on this page is legal advice.

Who should own AI compliance in a mid-sized company?

A person inside the company who knows its contracts, insurance and information, often the finance chief, the operations lead or the person who handles contracts, with a written half-page scope and a review date. The lawyer advises. The IT provider supports the tools but should not own the promises.

Does a vendor saying its tool is compliant make us compliant?

No. The statement is about the vendor's product under the vendor's conditions. Whether your use of it keeps the promises you made to customers, staff and your insurer is yours to show. Ask the vendor for its terms and where it keeps your data, and write the answer on your page.

What should I do if we already used AI in a way that broke a promise?

Stop that use, find out what was affected and write it down. Tell your lawyer the same day, and let your lawyer advise on whether anyone outside has to be told. Then fix the cause, which is usually a missing data rule or a missing owner, and date the change. This is general information, not legal advice.

How do I answer a customer's AI questionnaire?

With the page. Say what you use, who owns the question, what information stays out, what a person always checks and when it was last reviewed. Answer only what is true today, and say what is changing and by when. The page on who is responsible when AI makes a mistake at work covers the related question of what to write down.

Not sure where you stand? Take the 3-minute AI readiness scorecard.