Ireland's Data Protection Commission has ordered X to stop training its Grok models on third-party data gathered by its Grok Bot agent, and the company has complied. Data Protection Commissioner Dale Sunderland told Politico that the regulator raised "significant concerns" about the processing in recent weeks and asked X to suspend it. "We asked them to suspend the processing, and they've confirmed to us today that they've done that," he said. The confirmation came on Monday, November 16.
Grok Bot had been collecting this material since late October. The category at issue is not content users post about themselves, which has dominated AI training disputes for years, but personal information about people who never signed up for the product: photos a friend shares with an agent user, messages forwarded into a mailbox an agent reads, or personal data an agent scrapes from the open web. European privacy law treats that as third-party data, and controllers owe the people described in it information about the processing under Article 14 of the GDPR.
Sunderland was careful to separate the tool from the use. The commission does not object to AI agents as such, he said; the question is what happens when a provider moves into "further taking of that data and using it to further train the models." He added that the DPC is already talking to another, unnamed technology company about similar plans, before any processing has begun.
No fine has been issued and the commission has not published a formal decision, which makes this an early regulatory signal rather than a precedent with reasoning attached. The DPC enforces the GDPR for companies whose European headquarters are in Ireland, a list that includes Apple, Google, Meta, X and OpenAI, so its reading of a practice tends to shape what those providers offer across the European Union.
The timing sits awkwardly for X. Grok Bot is the company's push to turn the assistant from a chat interface into an agent that acts inside a user's accounts, and agentic products depend on ingesting exactly the kind of ambient personal data the commission just halted. Any provider wiring an assistant into mailboxes, shared drives or calendars inherits the same exposure, whether or not it operates in Europe.
The enforcement also extends a pattern from 2026 in which regulators have moved from the model's training corpus to the model's behaviour. Earlier actions in Europe focused on whether publicly posted material could be used for training. This one asks a different question: whether an autonomous system may gather information about bystanders in the course of doing its job, and whether that secondary collection can then become training material.
X has not commented publicly on the order. The commission has not said whether it will open a formal inquiry, and the identity of the second company in talks has not been disclosed. Both omissions matter for anyone building agents on European user data: the boundary set here applies to the collection step, not only to the training step, and it applies before a model is built rather than after.
Comments (0)
Log in to join the discussion
Log InNo comments yet