You may be using the system—but they may own the output.

In most AI vendor agreements, the agency assumes it owns everything: the data that goes in, the outputs that come out, and the models used for decision support. But in practice, many vendors retain the right to reuse your data to train their systems, withhold audit logs, or delay access to outputs. And that creates major downstream risk.

‍

What You Need to Clarify Before Signing

1. Who owns the inputs, outputs, and metadata?

Your contract should specify that your agency owns all system outputs—alerts, predictions, classifications, and logs—and can access and disclose them at any time.

2. Does the vendor retain training rights?

If your vendor is using your agency’s data to improve its models, that may introduce bias, increase liability, or make you responsible for decisions made elsewhere. Training rights should be opt-in and clearly defined.

3. Can the agency access the model logic or reasoning?

Even if the model is proprietary, your agency must be able to trace inputs to outputs—especially in the event of legal challenge or public inquiry.

4. What happens if the vendor goes offline?

Can you continue to access your data and logs? Can you migrate the system or outputs elsewhere? You need a continuity clause that accounts for vendor change, termination, or platform shutdown.