Can you use AI on work that is under an NDA?
By Matthew Eagle
Founder, Eagle AI Agency. Eight years as a staffing account executive.
Published . Updated .
A confidentiality agreement that never says the word AI can still rule out the tools you were about to use.
Usually yes, but the plan you are on decides it. Free tiers of AI tools commonly reserve the right to use what you submit to improve the product, and paid or business tiers commonly do not, which makes the tier you happen to be signed into a term of the confidentiality agreement you just signed. The reliable fix is to name the tools inside the agreement rather than leave it to be argued about later.
This came up for us this week. A new client sent over their own confidentiality agreement through a signing link, ready to accept in two clicks. Nothing hostile in it, nothing unusual. It also never used the word AI once, and it still changed how we were planning to do the work.
Why does a clause that never mentions AI still apply?
Because of what those clauses actually say. Most of them say you will not disclose the information to anyone else, and you will not use it for anything other than the work. Neither sentence has to name a technology in order to cover one.
Put a client's customer notes, candidate list or pricing into a tool that keeps a copy and uses it to get better, and you have arguably done both things at once. You have passed it to somebody else, and you have used it for something other than the job in front of you. Whether that is a breach is a question for a lawyer. Whether you want to find out the answer during a dispute is not.
What is actually different between a free plan and a paid one?
Often the data terms, and that is exactly the problem. The same product, from the same company, can give you a different answer on whether your input is used for training depending on which tier you are on and which settings are switched on inside it. Sometimes there is a toggle. Sometimes the toggle only exists above a certain tier.
So the honest answer to "do you put our data into AI" is not yes or no. It is: here are the tools, here is the tier, here is the setting, and here is what each one does with what we give it. That is a far better answer to be able to give. You can only give it if you have looked.
Should you name your tools in the contract?
We think yes, as an exhibit attached to the agreement rather than a sentence buried inside one. An exhibit does three jobs a clause cannot.
What a tool exhibit in the agreement does
| What the exhibit does | Why it matters |
|---|---|
| Names every tool that will touch the data | Turns "we are careful with your information" into something the client can check |
| Records the tier and the relevant setting for each one | The tier is the term, so it has to be written down somewhere, not remembered |
| Says how the list gets changed | You will add a tool eventually. Far better that it needs a notice than a re-signed agreement |
What happens when the tool list is wrong?
Ours was wrong on the first pass, and it is worth saying so plainly.
The first version of that exhibit described a setup neither side was actually on. It named the wrong plan for the client, the wrong plan for us, and a shared piece of infrastructure that was already scheduled to stop being shared. None of it was dishonest. It was written from memory instead of from the accounts.
It got caught because a second reader went through the draft cold, against reality rather than against the previous version. That is the check worth stealing. Reviewing a document against the last draft tells you what changed. Reviewing it against the world tells you whether it is true.
An exhibit that misdescribes your own stack is worse than no exhibit, because now the agreement says something false from the day it is signed, and you are the one who wrote it.
What do you do about a tool you are not allowed to use?
Take away the ability, not just the permission.
We ruled one tool out of this engagement entirely, because its free tier may use what is submitted to improve the model and the paid tier was not worth buying for this piece of work. Writing that down was the easy half. The half that matters is that the tool's access key came out of that client's working environment, so nobody can reach for it later on a busy afternoon and quietly break the agreement with the best of intentions.
A rule that depends on everybody remembering it is not a control. It is a hope.
What can you do before you sign, for free?
All of these cost nothing and take an afternoon.
Five free steps before you sign
- 1Find the training setting in every tool you use for client work, and turn it off.
- 2Put two-step verification on the mailbox the work runs through.
- 3Turn on device encryption on the machine the files sit on.
- 4Move client files out of any consumer sync folder, and keep them out of your code repository.
- 5Get anyone else who will touch the data under their own agreement before they touch it.
Do these first and the conversation changes completely. You stop defending a position and start describing a setup.
Questions people ask
Does this mean we cannot use AI on client work at all?
No, and that would be the wrong lesson. It means knowing which tools, on which plans, with which settings, and being willing to write it down. Most of our work runs through these tools. The point is that it runs through named ones.
What if the client never asks?
Most will not, which is exactly why offering it first lands so well. Handing over a clear list before anyone requests it is one of the cheapest trust signals available to a small company, and it costs you almost nothing.
Is a signed agreement enough on its own?
No. The agreement sets the rule and the setup enforces it. If a tool is off limits on paper but still sitting there logged in and one click away, you have documented an intention rather than built a control.
Who should be checking this, us or the client?
Whoever thinks of it first, and it should be you. The client signing an agreement with you is rarely in a position to know what your stack is. You are.
If you are running a small team and this is one more thing you do not have time to work out, that is the normal situation rather than a failure. Book a call and we can walk through what you are actually running.
More notes
- What stops AI outreach from sending the wrong message?This week a prospect received our AI's private note about them as the opening message of a sequence. Here is how it got through, how we made sure it was the only one, and what checks the words now.
- Why do verified emails still bounce?This week a list that passed two separate email checks still bounced hard enough to stop its own campaign. Here is what the checks missed, and what we now run before every send.
- How do you move to a new CRM without losing your history?We moved a small firm's years of CRM history into a new system and checked every record. The next day they told us the notes had not come over. They had.
- Why did the link in your email open a warning page?We approved an email with a clean payment link. Our customer clicked it and landed on a warning page instead.
- Is your outreach broken, or is the benchmark wrong?The fastest way to wreck a campaign that is working is to measure it against a number that came from somewhere else.
- Do more pages actually help your website rank?The advice to publish more assumes pages are the unit of measurement. They are not, and a site full of near identical pages can be worth less than a site with a handful of good ones.
- How do you check if your cold email is going to spam?We spent days proving our email was landing in spam. Every test was pointed at a kind of mailbox almost nobody on the list used.
- Should a staffing firm hire a recruiter or rent a desk?Most small staffing firms are not short of a recruiter. They are short of a recruiter they can afford to carry while one ramps.
- Why does AI make things up, and how do you stop it?A model handed something unusable does not stop. It produces something plausible, and it will produce a different plausible thing the next time you ask.
- What breaks when you automate a process nobody is watching?Automation rarely fails loudly. It fails quietly, and keeps reporting that it is fine.
- Why did your cold outreach run out of leads?Usually it is not the message. You have simply run out of people to send it to.
- Do you need to hire a salesperson, or a system?The honest answer depends on one thing, and it is not budget.
- What do you do about sales when you are already at full capacity?The trap is not that you are too busy. It is that being busy feels like being safe.
Want the version of this for your business?
Twenty minutes, your actual situation, a straight answer about what we would do first.
20 minutes. You leave with a straight answer.