• May 26, 2026

A Checklist for Higher Ed Institutions Reviewing AI Contracts

A pen on a sheet of paper with lines for signature and date.

A Checklist for Higher Ed Institutions Reviewing AI Contracts

A Checklist for Higher Ed Institutions Reviewing AI Contracts 1024 536 Vantage Technology Consulting Group

A pen on a sheet of paper with lines for signature and date.

Colleges and universities have survived the printing press, the Internet, and the learning management system*. Now comes artificial intelligence (AI) and a fresh stack of vendor contracts that your legal team, IT department, and provost’s office are all sure that someone else reviewed.

But are they reviewing those contracts for the right things? AI contracts are not like traditional software agreements. The stakes are different. Your students’ work could train a commercial model. Your researchers’ data could live on servers in jurisdictions you have to look up on a map. And that “flexible pricing” clause might let the vendor swap out the underlying model without telling you first.

When we help our higher education clients evaluate vendor products with embedded AI, we go straight to the contract, not the sales materials. The contract is where vendors define how AI is used in their products and what they commit to regarding that use. It is also where the cybersecurity and privacy risks that matter most to institutions and their campus communities either get addressed or are quietly ignored.

10 Things We Look for When Reviewing AI Terms in Vendor Contracts

This checklist reflects what we have learned working alongside higher education institutions as they navigate AI procurement. We’ve included some of the common terms that show up, the risks that get missed, and the questions worth asking. Because we are technology consultants, not lawyers, our list should be considered informed guidance, not legal advice. If in doubt, the best course of action is to bring in counsel to decipher complicated contractual clauses†.

  1. Student and faculty data cannot be used to train vendor models.
    Why it matters: FERPA protects student education records, and using those records to improve a commercial product almost certainly violates it. Research data may carry its own protections that need to be considered. If the contract says the vendor can use “usage data for model improvement,” you are giving a big (big) gift to the vendor.
  2. You own the outputs.
    Why it matters: When a faculty member uses a licensed AI tool to draft a report, a grant proposal, or course materials, who owns the result? The answer should be the institution or the user. If the contract is silent on ownership, or assigns ownership of outputs to the vendor, you have an opportunity to negotiate. Insist on explicit assignment of output ownership to the institution or the user.
  3. The underlying model cannot change without notice.
    Why it matters: You evaluated and approved Tool A. If the vendor silently swaps in Tool B six months later, your compliance reviews, your IRB approvals, and your trust in the outputs may all be void. A different model with different behaviors may need a new review. Require advance written notice of material model changes.
  4. Custom models built on your data belong to you.
    Why it matters: Some AI vendors offer fine-tuning or model customization using your institution’s data. If that custom model becomes valuable—and it very well might—who owns it? The answer should be you. If the contract says otherwise, you have an opportunity to negotiate.
  5. All sub-processors that touch your data are disclosed.
    Why it matters: “We use AWS” is not a sub-processor list. Many AI vendors route data through multiple third parties like model hosts, annotation vendors, or cloud regions. You are entitled to know exactly who touches your data and where.
  6. Data residency is specified.
    Why it matters: Research data, especially federally funded research, often cannot leave the country. Student data under FERPA have similar sensitivities. “Stored securely in the cloud” is not a data residency policy. Get a country and, ideally, a region so that you understand clearly where your data will be stored.
  7. Data is deleted after contract termination.
    Why it matters: When you leave a vendor, your data should leave with you, or at least be securely destroyed. Vague language like “we may retain data for operational purposes” gives the vendor indefinite access after you have stopped being a customer. Specify a deletion timeline and require written certification of that deletion.
  8. Liability for AI-related harm is not excluded.
    Why it matters: Many vendors try to exclude AI outputs from their standard liability cap. That means if a hallucinated financial aid calculation harms a student, or a flawed research summary causes downstream problems, the institution bears the cost of the mistake. Push back on any clause that treats AI outputs as outside of the vendor’s responsibility.
  9. Human review is required for high-stakes decisions.
    Why it matters: Admissions, financial aid, academic integrity determinations, and employee performance reviews are all high-stakes decisions where mistakes have real-world consequences for a specific person. If AI touches any of these, the contract (and your internal policy) should require a human in the loop. Some AI laws are even starting to require this, so it is best to be mindful of the issue during contract negotiations.
  10. Applicable AI regulations are identified, and compliance obligations are assigned.
    Why it matters: Governments worldwide are passing AI-specific legislation, and several classify certain uses in education, particularly those affecting student assessments or admissions, as high-risk transactions with high-risk data. Knowing what these regulations are is especially important if your institution operates or serves students from other countries. The contract should specify which regulations apply to your use case and who is responsible for compliance. This is not a question to leave unanswered until there is a problem. Instead, you want to discuss this during contract negotiation.

Get It In Writing

Vendor assurances are not contract terms. The only protection that matters is the language in the document you sign. Before signing any contract, ask your vendor to identify the contractual terms that are most important to you. I like to start with #1 on our list: “Can you show me where in this contract you commit to not training on our data?”

If they can answer quickly and point to specific language, that is a positive sign. If they pause, check with legal, or send you a blog post instead of contract language, that is a good indication that you want to slow down and tread carefully in your review of the product and contract.

When a vendor does point you to specific contract language, take the time to read it carefully. Contract terms can be narrow in ways that matter. A clause that protects “personal data” may not cover research data. A promise to “anonymize” inputs is not the same as a promise not to use them in the first place. If something is unclear, do not guess. Instead elevate your questions to campus legal counsel‡.

AI technologies are evolving fast, and vendor contracts are not always keeping pace. But the effort to get these contracts right is worth it. Your institution carries real obligations to its students, faculty, staff, and research communities. A contract that reflects those obligations is not a bureaucratic hurdle. Rather, it is part of how you protect the people who trust you.

This post was authored by Senior Principal and Partner Joanna Lyn Grama, JD, CISSP, who is a lawyer, but not your institution’s lawyer. Joanna leads Vantage’s information security practice and works with clients to examine and improve their cybersecurity, data privacy programs, technology governance, and compliance.

*I realize that the learning management system has been the talk of the town recently with the May 2026 Canvas incident.

†This is not just a disclaimer. Bringing in the lawyers when reviewing complicated contracts is genuinely good life advice.

‡Just to reiterate, involving legal counsel to interpret contract language is never the wrong call.