On March 17, Vercel updated its Terms of Service: the new terms allow AI training on customer data, on an optional basis, with the corresponding opt-out window closing March 31. Today is March 25, which leaves six days.
Updates like this usually sit quietly in a changelog, but the question it touches is core data governance: whether content hosted on a platform can be used to train models is now something every team must answer actively — and silence is no longer a neutral position.
The Three Things That Changed
- Timing: the new terms took effect March 17; the opt-out window closes March 31
- Substance: customer data may be used for “optional” AI training — the default stance is defined by the terms, participation is decided by the user
- Action: teams that do not intend to participate must complete the opt-out within the window, not at the next convenient deploy
Note the semantics of “optional”: the mechanism exists and the choice belongs to the user, but the default behavior is defined by the document’s text. When reading terms like these, the productive question is what happens if you do nothing — not what happens if you act.
Why This Is Not a Small Thing
For a personal project this may be a privacy preference. For a team with compliance obligations, the weight is entirely different. Data-handling rules in regulated industries, data clauses signed with end customers, and cross-border transfer restrictions may all forbid the premise that “the platform may use our data for model training.” The terms hand the decision to the user — and the burden of proof along with it: no statement within the window, and the consequences are yours.
The timeline compounds it. Terms effective March 17, opt-out closing March 31: a two-week window, announced in a changelog, compresses real reaction time to days — and for organizations that need legal sign-off, that practically guarantees a scramble against the deadline.
The deeper issue is process. Most teams learned about this from a changelog or a community thread. A vendor terms change that never enters procurement and legal tracking means vendor management still has a hole in daily operations. Vercel is not the first company to move this line, which is exactly why the reaction playbook matters more than any single vendor.
A Six-Day Checklist
- Confirm the opt-out status at both the account and organization level; personal accounts and company orgs need separate checks
- Inventory any contracts or regulations that prohibit data use for model training; for projects in doubt, opt out first and discuss later
- Add “AI training clauses” to the vendor evaluation checklist, so the next change is caught by legal first — not scrolled past by a developer
If your organization has no single owner for platform terms, this week is the argument for creating one.
Six days is not long, but it is enough to walk through the required approvals. Missing the deadline may not show an immediate cost; it tends to resurface at the next audit or renewal negotiation, when the leverage is gone. Arguing about the fairness of the terms after the window closes will only cost more.
Sources
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.