Before you hand over a single file, make sure the vendor sitting across the table has earned your trust.
Here's what to ask before you sign anything with an AI vendor: where your data physically lives, who inside their company can see it, what happens to it the day the contract ends, and what proof they can show you instead of just telling you.
This is a vendor-vetting framework built for CEOs, COOs, and operational leaders who are personally accountable for how their organization's data gets used, whether they run the vetting themselves or hand it to someone on their team. Get concrete answers to these questions in writing before any data changes hands.
Most disaster stories follow the same pattern: a leader skipped a question they assumed was obvious. They trusted a vendor's word instead of demanding proof, and found out the hard way that "we take security seriously" means nothing without specifics behind it.
Ask Where Your Data Actually Lives
Start with the most basic question in the entire vetting process: Where does the data go once you hand it over? Not the marketing answer. The actual answer, with names of servers, regions, and storage providers attached.
Vendors love to say "the cloud" like that settles the matter. It doesn't. The cloud is a business decision made by someone else, on infrastructure you've never inspected, governed by laws you may not have read. Data stored in one country carries different legal protections than data stored in another.
Which data centers hold our information?
Ask whether the vendor stores data with a major provider like AWS, Google Cloud, or Microsoft Azure, or runs its own servers. If they run their own, ask who manages security patches and how often. A vendor who can't answer quickly hasn't thought about it seriously, and a vendor who hasn't thought about it seriously shouldn't be trusted with anything sensitive.
Does our data get used to train your AI models?
This question trips up more vendors than any other, because the honest answer often reveals that your proprietary information could end up shaping outputs for other clients, including your competitors. Some vendors default to using client data for model training unless you explicitly opt out. Others build walls between clients by design. You need to know which one you're dealing with before anything gets uploaded.
Do any regulations dictate where our data can legally live?
If you operate in healthcare, finance, or government contracting, rules like HIPAA or GDPR may dictate exactly where certain data can and cannot reside. A vendor unfamiliar with these requirements hasn't done business with companies like yours before, and that inexperience becomes your risk the moment you sign.
Get every answer in writing. Verbal reassurance evaporates the moment something goes wrong.
Ask Who Else Gets to See It
Every vendor has a team. Every team has access. The question you need answered is how wide that access actually spreads before your data ever reaches a screen.
The stakes here are higher than most leaders assume. Research on AI use in the workplace shows that most employees use AI tools their employer never approved, and a majority have uploaded sensitive company data into those tools without authorization. That same pattern of casual, unmonitored access happens inside vendor organizations too, unless the vendor has built specific controls to prevent it.
How many people at your company can view our raw data?
Some vendors restrict access to a small, vetted group. Others let dozens of people touch client information as part of daily operations. Neither answer is automatically disqualifying on its own, but you need to know which situation you're stepping into.
A vendor with loose internal access controls and no compensating safeguards is one careless employee away from a leak. A vendor with broad access paired with strong monitoring, logging, and training may be perfectly safe. The access number alone doesn't tell you the answer. What the vendor does around that number does.
What third parties or subcontractors touch our data?
Many AI vendors plug in third-party tools for storage, analytics, or model hosting. Each third party in that chain is another door your data passes through. Request a full list of subprocessors and what each one touches.
What happens to an employee's access when they leave the company?
Ask whether access gets revoked immediately or lingers. A vendor without a clear offboarding process for their own staff is unlikely to have a clear one for anything else either. Offshore support and outsourced teams follow the same logic: they aren't automatically a risk, but they need documented access controls behind them, not just a lower price tag.
Trust is built on visibility. If a vendor hesitates to show you who touches your data, that hesitation is the answer.
Ask What Happens When the Contract Ends
Most leaders focus every ounce of attention on the beginning of a vendor relationship and almost none on the end. That's backwards. The exit terms tell you more about a vendor's character than the sales pitch ever will.
What happens to our data the day the contract terminates?
Does it get deleted? Archived? Does it sit on a server somewhere indefinitely because nobody wrote a policy for it? Vendors who haven't thought this through usually default to keeping everything, because deleting data requires effort and nobody built that effort into their process.
Can you give us a specific deletion timeline in writing?
"We'll handle it" is not a timeline. You want a number of days, a named process, and a way to confirm deletion actually happened, such as a certificate of destruction.
What happens to backups, and to any model trained on our data?
Live data might get deleted on schedule while backup copies sit untouched for months or years afterward. A vendor with a real data governance program will have a policy covering backups specifically.
And if the vendor built or fine-tuned a model using your information, ask directly whether that model gets destroyed too, or whether it lives on serving other clients with your fingerprints inside it. This question rarely gets asked, and it should be one of the first.
A vendor confident in their practices will answer all of this without flinching. A vendor who stalls just gave you the only warning you needed.
Ask How They Prove It, Not Just Promise It
Every vendor will tell you their security is strong. None of that means anything without documentation behind it.
Can you produce a SOC 2 Type II report?
This is the standard most legitimate vendors can produce, and it covers how they handle security, availability, and confidentiality over time, not just at a single snapshot. A vendor who treats this request as unusual is still building their security program while working with your data, not before.
Are you ISO 27001 certified, or do you follow an equivalent standard?
Especially for vendors operating internationally, this signals a formal, audited information security management system rather than a collection of good intentions. Certification isn't everything, but the absence of any recognized standard should raise your guard.
When was your last penetration test, and what did it find?
A vendor serious about security pays outside experts to try to break in, then fixes what they find. A vendor who can't remember their last test hasn't made this a priority, regardless of what their website claims.
What is your incident response plan?
You want to know who gets notified, how fast, and what happens in the first hour after something goes wrong. Vague answers here are the clearest warning sign on this entire list. Shadow AI incidents, where unapproved tools expose company data, already cost companies hundreds of thousands of dollars more per breach than standard incidents. A vendor without a real response plan turns that number into your number.
Ask for references from current clients in your industry, then actually call them. A vendor confident in their track record will connect you without hesitation.
Documentation replaces trust with proof. Any vendor unwilling to produce it has already told you everything you need to know.
None of this means every imperfect answer disqualifies a vendor. A vendor with broad internal access isn't automatically unsafe if they pair it with real monitoring and training. Just like a vendor using offshore support isn't automatically risky if they've built documented controls around it. The goal is to find a vendor who can answer specifically, back it up in writing, and show you they've actually thought about the question before you asked it.
What This Looks Like in Practice
Picture walking into a vendor relationship where you already know where your data lives, who touches it, what happens when the contract ends, and how the vendor proves their claims instead of just stating them. Board meetings get easier because you can answer the hard questions before anyone asks them. Headlines about data leaks at other companies stop triggering that familiar drop in your stomach, because you already did the work.
That confidence comes from asking the right questions before you sign, not from hoping the vendor has already answered them somewhere in the fine print.
If you want a structured way to build this kind of vetting discipline across your whole leadership team, not just for one vendor decision, our AI Mastery for Business Leaders course walks your team through exactly how to vet AI vendors and trainers, protect company data, and lead AI initiatives with real confidence. Enroll now.