How do I check whether a tool may train AI on my files?
Nobody writes "we will train our model on your client files". The sentence that does the work is quieter, and it is usually about improving the service.
Start with the clause where you give the company a licence to your content. Every service needs one, simply to store your files and show them back to you. What matters is what the licence is for. "To provide the service" is the narrow version. Words like "improve", "develop new features", "analytics" or "machine learning" widen it. When SoundCloud's terms were found in 2025 to allow AI training on what users uploaded, the wording had been added without an announcement. When WeTransfer changed its terms that summer, the phrase that set people off was about improving machine learning models. Both companies ended up explaining themselves in public.
Then check which agreement you are actually under. Many companies keep one set of terms for free and personal accounts and another for paid business plans, and the promises about your content can differ between them. A practice that signed up years ago on someone's personal account may be living under the weaker set without knowing it.
Then look for a switch. Some services let you turn training off in the settings, and some have it on until you do. If the terms mention a choice, go and find it rather than assuming how it is set.
Then read the data processing agreement, if there is one. It is the document where a company says what it will do with personal data on your behalf, and it is often plainer than the terms.
If after all that you still cannot tell, ask in writing. A short email to the vendor's privacy address, asking whether customer content on your plan is used to train models, costs you two minutes. The answer is worth keeping. So is the absence of one.
You are allowed to find all this tedious. It is. It is also the difference between a tool you use and a tool that uses you back.
Last checked 16 November 2026. Information, not legal advice.