Summary: Vercel revised its Terms of Service on March 17, adding language that lets customer data feed AI training. Yet it lands on the foundation of data governance: whether material hosted on a platform may end up training models has become a question every team has to answer out loud. The arrangement is optional, and the door out — the opt-out window — shuts on March 31.
Vercel revised its Terms of Service on March 17, adding language that lets customer data feed AI training. The arrangement is optional, and the door out — the opt-out window — shuts on March 31. With today being March 25, six days remain.
By habit, a change like this would live and die inside one changelog entry. Yet it lands on the foundation of data governance: whether material hosted on a platform may end up training models has become a question every team has to answer out loud. Staying quiet no longer counts as taking no side.
What the Update Actually Contains
- When: the terms went live on March 17, and the opt-out must be completed before March 31
- What: customer data can enter “optional” AI training — the terms set the default behavior, while the user holds the decision to take part
- To do: teams declining to participate have to finish the opt-out inside the window, rather than deferring it to whatever deploy comes next
The word “optional” deserves a second look: it confirms the mechanism is there and the choice sits with the user, yet whatever happens by default is settled by the text itself. So the first question to ask of terms like these is what occurs when you refrain — not what occurs when you act.
The Stakes Are Heavier Than They Look
A personal project might treat this as one more privacy toggle. A team carrying compliance obligations weighs it on a different scale entirely. Data-handling requirements in regulated sectors, the data clauses committed to end customers, and limits on cross-border transfers may all rule out the premise that “the platform may train models on our data.” While handing the decision to users, the terms also pass them the burden of proof: say nothing before the window closes, and what follows is on you.
The calendar sharpens it further. Terms take effect March 17, opt-out closes March 31 — a two-week gap, announced in a changelog, shrinks the workable response to days. For organizations that route decisions through legal sign-off, someone sprinting toward the deadline is close to inevitable.
Beneath that sits a process problem. Word reached most teams through a changelog entry or a community thread. When a vendor rewrites its terms and neither procurement nor legal ever logs it, vendor management plainly has a gap in its routine. Vercel is hardly the first company to redraw this particular line — which is the strongest argument for fixing the playbook rather than fixating on one vendor.
What to Finish Before the Deadline
- Check the opt-out state at the account level and the organization level alike; personal accounts and company orgs must each be verified on their own
- List every contract and regulation that bars data from being used in model training; projects carrying any doubt should exit first and argue later
- Put “AI training clauses” on the vendor evaluation checklist, so legal reads the next change first instead of a developer scrolling past it
And if nobody in your organization currently owns platform terms, this week makes the case for appointing an owner.
Six days is tight, yet it will carry the necessary approvals from start to finish. The price of a missed deadline rarely surfaces right away; it waits for the next audit or renewal negotiation, arriving after your leverage has already left the room. Once the window has shut, arguing over whether the terms were fair only makes the bill larger — finish the action instead.