Wrivio
Get Wrivio
5 min readBy Wrivio Team

Can AI Train on Your Data Under Legitimate Interest? The Proposed GDPR Change

One of the more consequential ideas in the EU’s 2026 Digital Omnibus proposal was to recognize that AI providers may process personal data to train and operate their systems on the basis of “legitimate interest,” with safeguards expected to accompany it. The phrase sounds technical, but it decides something concrete: whether a company can use personal data to train an AI model without first asking the people it belongs to. Here is what it means and what the change would and would not allow.

What “Legitimate Interest” Is

The GDPR gives several lawful bases for processing personal data. Consent is one; “legitimate interest” is another. Legitimate interest lets an organization process data without consent when it has a genuine need, that need is not overridden by the individual’s rights, and safeguards are in place. It is a balancing test, not a free pass, and it already covers a lot of ordinary processing. The GDPR text is at gdpr-info.eu. We set out how these obligations sit alongside the AI Act in GDPR and AI Act parallel enforcement.

The proposal would make explicit that training and running AI systems can qualify, which removes a legal grey area that has hung over model training for years.

Why It Is Contested

The reason this is a fight, and why the EDPB and EDPS pushed back, covered in Europe’s privacy watchdogs pushed back on weaker GDPR, is that “legitimate interest” for AI training shifts the default. Under consent, your data is not used unless you say yes. Under legitimate interest, it may be used unless you object, assuming the balancing test and safeguards hold. For personal data swept into a training set, that is a meaningful change in who has to act.

The safeguards matter enormously and were not fully settled: the right to object, data minimisation, and limits on special-category data are the difference between a workable basis and a loophole.

What It Means For Your Own Writing

For you as a writer, the direct effect is about what happens to text you put into AI tools. If training on personal data under legitimate interest becomes the norm, then anything you paste into a consumer tool is more likely, not less, to end up shaping a future model unless you have opted out or the tool commits not to train. That raises the value of knowing a tool’s training and retention terms, covered in questions to ask an AI vendor about data and what zero data retention actually means.

Before:

If legitimate interest covers AI training, our client data is fine to use in any AI tool.

After:

The proposal would let providers train on personal data under legitimate interest, subject to safeguards and a right to object, and it is still contested. That makes it more likely, not less, that data pasted into consumer tools feeds a model. We will keep client data in tools with explicit no-training terms, or out of third-party tools entirely.

The second version reads the change as a reason for more care, which is the correct direction.

The Hedge That Does Not Rely On The Balancing Test

The cleanest way to keep your text out of a training set is to not transmit it. Text rewritten locally on your machine is never sent to a provider, so no lawful basis, consent or legitimate interest, applies to it, because no processing by a third party happens. For confidential and personal material, that removes the entire question. We compare the approaches in local AI versus cloud AI for confidential writing.

A Wrivio Context for a data-basis note could say:

Rewrite this as a precise internal note. Keep every legal term, body name, and date exactly as written. Distinguish clearly between consent and legitimate interest and between proposed and settled. Do not state a proposal as current law. Flag anything to confirm with counsel.

Press Ctrl+Shift+Space, paste your draft, and check the diff. A rewrite that keeps “proposed” and the safeguards intact is doing its job. This is a summary, not legal advice.

Common Questions

What is the proposed GDPR change for AI training?

The 2026 Digital Omnibus floated recognizing that AI providers may process personal data to train and operate systems on the basis of legitimate interest, with safeguards expected. It is a proposal, contested by EU privacy regulators, not settled law.

What is legitimate interest under GDPR?

A lawful basis for processing personal data without consent when an organization has a genuine need, that need is not overridden by the individual’s rights, and safeguards apply. It is a balancing test, not an automatic permission.

Does this mean my data can be used to train AI without asking?

If the change holds with its safeguards, personal data could be used under legitimate interest subject to a right to object, rather than requiring prior consent. That is why it is contested, and why knowing a tool’s training terms matters.

How do I keep my text out of AI training sets?

Use tools with explicit no-training commitments, exercise opt-outs where offered, and for confidential material draft locally so the text is never transmitted to a provider in the first place.

Download Wrivio for Windows to rewrite personal and confidential text locally, where no provider processes it at all.