Generator
Data Boundary Clause Generator
Procurement blocks more AI deals than pricing does, and usually over four questions: do you train on our data, how long do you keep it, who else touches it, and what happens when we leave. This drafts those clauses.
- Four clauses that matterThe ones that block deals
- Both directionsAsk them, or answer them
- Not legal adviceA starting draft
Work it out
Your numbers
Your system prompt
Drafting aid only. These clauses are deliberately plain rather than exhaustive, so that both sides can read them. Have counsel review before signature.
What the numbers mean
Derived embeddings are the clause people forget
Vendors sometimes delete conversations but keep the vectors built from them, on the basis that a vector is not the content. Name embeddings explicitly or you have not asked for deletion.
A silent model swap is a change of processing
If the vendor changes which provider runs inference, your data now goes somewhere else. That deserves notice, and most standard terms do not give it.
Ask for the report, not the badge
A certification logo on a marketing page is not evidence. The clause that matters is the one requiring the actual report on request.
Plain clauses get signed faster
Exhaustive drafting invites redlines. Four clear obligations that both sides can read usually close quicker than fourteen that need a lawyer to parse.
Questions
Treat them as a starting draft that makes the conversation concrete. They are deliberately plain, and your counsel should review before signature.
That is useful information early. Some vendors genuinely cannot offer it on their lower tiers, and it is better to know before integration than after.
does not train on submitted data. Retention periods, the subprocessor list and a signed DPA are planned before anyone is charged — the security page states what is settled and what is not.
Want this run against your actual site?
Tell us the page your assistant lives on and the questions you care about. A person runs them and sends back the raw answers.