What Should We Refuse to Sign? Five AI Contract Terms Decided Once, at Signature
Data ownership, training rights, exit, price, and liability are settled the day you sign and renegotiated almost never. In a clinic one of them is not just commercial, it is a privacy obligation you already hold. Here is what each clause decides, what the standard wording quietly grants, and the version worth asking for instead.
What this settles
Which clauses in a clinic AI vendor's agreement you will not accept as written, decided before your practice is dependent on the product.
In this article
Every other decision in this series can be revisited. You can change what you automate, replace an owner, stop a project. The contract is the exception. Five of its clauses are decided the day you sign, and the moment you sign them your leverage transfers to the other side, permanently, because from then on you are the one who would be inconvenienced by leaving.
That asymmetry is the whole reason this is a decision a practice owner makes rather than something forwarded to whoever handles the paperwork.
The five that matter
Ownership is usually fine. Liability is usually one-sided and, awkwardly, usually normal. The two that quietly cost the most are the two nobody reads: training rights and price. In healthcare the first of those stops being a commercial question and becomes a privacy one.
A perpetual licence to use your content to improve the service sounds administrative. What it grants is the right to learn from your notes, your templates, and your recorded patient conversations, and to sell the resulting product to anyone, including the clinic down the road. Outside healthcare that is a trade some businesses make deliberately for a discount. Here it is different: the material was collected to provide care, improving a vendor product is not that purpose, and no discount makes that reasoning go away.
Fees amendable on thirty days notice is the other one. It reads as boilerplate until the moment your clinicians genuinely depend on the tool and would revolt at losing it, which is precisely when a price rise is least refusable. The clause is not unfair so much as unbalanced against a buyer who will, by design, be more dependent later than they are today.
The clause you get, and the clause to ask for
Most of this is negotiable in ways buyers assume it is not, particularly with smaller vendors who would rather adjust a schedule than lose a customer.
”Customer Content is never used to train models serving other customers. Vendor acts as Customer’s agent under PHIPA s.17. Fees are fixed for the initial term and may rise once per year thereafter, capped, with ninety days notice.”
Same product, same price today. The difference only shows up in year two.
”Customer grants a perpetual, irrevocable licence to use Customer Content to improve the Services. Fees may be amended upon thirty days notice.”
Reads like boilerplate. Grants your patients’ encounters forever, and your budget annually.
Neither version costs the vendor anything today, which is exactly why asking works more often than buyers expect. A vendor who refuses both has told you something useful about how they intend to make money from the relationship.
Three honest outcomes
You have read the five. Now what?
The agreement is in front of you and two clauses are worse than you would like. What actually happens next?
Sign as written
Fine for a small, replaceable tool that never touches patient data. Know that you chose it.
Counter the two
Training rights and a price cap. Costs an email, works surprisingly often, and the answer is informative either way.
Walk
Reserved for a vendor who will not put data handling in writing, or will not sign as your agent. That refusal is the finding.
The instinct to walk away over an imperfect clause is usually wrong, and so is signing without reading. Countering two specific clauses is a fifteen-minute email that either improves the deal or reveals a vendor whose business model depends on terms they will not discuss.
Exit is the clause that ages worst
Of the five, the one buyers regret most is the one that looks most harmless. A right to export your data means very little without a format and a deadline attached, and “you may request an export of your data” is compatible with receiving a PDF of something that used to be structured, eight weeks later. For a clinic there is a second edge to this: those records are part of your patients’ health record, and your obligation to produce them on request does not pause while a vendor takes eight weeks.
This is the contractual half of what happens when the model you built on changes. The technical half is keeping your integration thin. The contractual half is making sure the thing you would be migrating is legally and practically portable on a stated timeline. Neither works without the other.
Every one of these five is free to ask for before signature and expensive to request after it. The signature is the leverage.
What to do this week
Open the agreement for the AI tool your practice already leans on hardest and search it for five words: own, train, terminate, fees, and agent. Read only those clauses, which takes about fifteen minutes. If the terms are worse than you assumed, the renewal date is your next negotiation, and knowing that now is the difference between a negotiation and a formality.
About to sign something that touches patient data? Book a free consult and we will read the five clauses with you, check the PHIPA agency wording, and mark the two worth countering before you sign. See more on Second Opinion.
The decision brief
One page for the person you are going to forward this to.
- The decision
- Which clauses in a clinic AI vendor's agreement you will not accept as written, decided before your practice is dependent on the product.
- What getting it wrong costs
- You discover the terms at the worst moment: a price rise you have no leverage to refuse, an export of your notes that arrives as a useless PDF dump, or the discovery that your patients' encounters improved a product sold to the clinic down the road. None of these are recoverable after signature, because the only leverage you ever had was the signature itself.
- What you can do this week
- Open the agreement for the AI tool you already depend on most and search it for four words: own, train, terminate, and fees. Read only those clauses. Fifteen minutes tells you what you actually agreed to, and whether the renewal is a negotiation or a formality.
- What to tell your team
- Before we sign anything that touches patient data, we check five terms: who owns it, whether they train on it, what we can take when we leave, how prices can change, and who is liable when it is wrong. And they sign as our agent under PHIPA, or we do not sign at all.
Frequently asked questions
What should I look for in an AI vendor contract?
Five clauses carry nearly all the risk: who owns the data you put in, whether the vendor may train on it, what you can export when you leave and in what format, how and how often prices can change, and who is liable when the output is wrong. In Ontario there is a sixth that is not negotiable at all: the vendor has to accept being your agent under section 17 of PHIPA. Most agreements are reasonable on ownership and quietly generous to the vendor on training rights and price changes, which are the two that cost the most later.
Should I let a vendor train on my clinic's data?
No, and in healthcare this is not the priced concession it is in other industries. Elsewhere training rights are a commercial trade you can make deliberately in exchange for a discount. Here the material is personal health information collected for the purpose of providing care, and improving a vendor's model is not that purpose. Treat any training clause touching patient data as a refusal rather than a negotiation, and be suspicious of a discount offered for it.
What is a fair exit clause for an AI tool?
One that names the format and the deadline. A right to export means little if the export is a PDF dump of something that used to be structured, or if it arrives six weeks after you asked. Ask for your data in a machine-readable format, within a stated number of days, including anything the system generated on your behalf. A vendor confident in their product rarely objects.
Who is liable when an AI tool gives a patient wrong information?
In almost every standard agreement, you are, and the vendor's liability is capped at what you paid them. In a clinic that allocation is even starker, because the privacy obligation and the clinical accountability were always yours and no contract moves them. That is not by itself a reason to walk away, but it should change your design: if the risk sits with you regardless, then human review on anything a patient reads or a chart receives is not caution, it is the only part of the arrangement you actually control.
Before You Spend
All 10 decisionsThe decisions an owner or CEO makes before the first dollar of AI budget leaves the building. One decision per article, each ending in a page you can put in front of your leadership team.
Found this useful?
Share this with your network on LinkedIn, it helps more than you think.
Enjoyed this read? Get the next one in your inbox.
When there is something worth your time, you will be first to know. No spam, unsubscribe anytime.
Keep reading
What Are We Exposed To, and What Is Our Policy? The One Page Every Practice Owner Should Write
Your staff are already using AI, so the policy question is not whether to allow it. It is whether the rules exist in writing or in each person's private judgement. Exposure comes in three shapes, and in a clinic the one most policies address is not the one most likely to reach a patient.
Read article
Agent Washing: How to Tell a Real AI Agent From an Automation With a New Sticker
Every clinic product renamed itself an agent this year, and the word stopped carrying information. A five-scenario quiz trains your eye, and five questions expose what a vendor actually built, because the label decides the price you pay, the way it fails, and the oversight your practice owes it.
Read article
Shadow AI in a Clinic: The Adoption Decision Your Staff Already Made For You
Your staff already made the call on AI. Not you, not a partners' meeting, one paste into a personal chat window at a time. In a clinic each of those pastes is an unlogged disclosure of personal health information. A short exposure check shows where you stand, and the fix is smaller than a policy document.
Read articleHave software that should be smarter?
Let’s map a free AI-transformation roadmap for your product.