Can I delete my data after AI trains on it?
The short answer
You can ask, but deleting files does not undo training. Troveo says trained models survive the license ending because data cannot be untrained. An FTC order announced in February 2024 made Avast delete derived data and algorithms, but that involved a vendor, not a company licensing its own archive.
Three questions hide inside this one. Can the buyer keep the files? Can the buyer keep the model? Can a regulator reach the model? The sources answer each one differently.
What do the companies say about deleting files?
micro1's data partnerships page says data is kept for a limited time under agreed retention policies. On request, or at the end of an engagement, it is deleted under the agreed terms. Troveo's guide lists whether data must be deleted at expiry among the terms to settle.
These are self-reported statements. The contract is what binds.
Does deleting undo training?
Troveo says no. Its guide says buyers typically insist that models already trained survive the license ending, "because nothing can be untrained."
So I'd split the request into three separate asks:
- stopping future access
- deleting a retained copy
- changing a model already built using the records
Can a regulator order a model deleted?
It has happened in FTC orders. The FTC order against Avast, announced 22 February 2024, reached any algorithms Jumpshot built from browsing data the FTC says Avast sold against its promises to users.
It's a warning for any company that passes data on, and weak precedent for a company licensing its own archive.
The other orders, and what model deletion does and doesn't reach, are in our page on FTC model deletion orders.
Does the CCPA let me stop a buyer's misuse after it trains on my data?
California's sale statute (CCPA text) requires a seller of personal information to hold rights in the buyer agreement. The seller must be able to take reasonable steps to stop and fix unauthorized use. Section 1798.100(d) lists that right. It reaches use of the data. The statute does not say it reaches a trained model.
What should I look for in a draft?
In a draft, I'd look for five things:
- the deletion trigger
- the deadline
- whether backups and derived data are included
- whether you get written confirmation
- whether the buyer's rights over models survive the term
The buyers page records what each buyer publishes about retention.
micro1 may pay us a referral fee if a company we introduce goes ahead with it. How we are paid
The managed readiness service lines up the deletion terms from each draft for your lawyer.
Talk to a privacy or commercial lawyer in your state before you sign.
Vaultfish helps you run the process. We are not lawyers and this is not legal advice. If you need counsel who knows this area, we can point you to one. Ask for a referral
