Ireland's Data Protection Commission (DPC) has pressured X into halting the training of its Grok AI models on third-party data, according to a report from Heise.

The DPC is the lead data-protection authority for X in the European Union, because the company's European operations are based in Ireland. The regulator's concern is material that X does not collect directly from its own users — third-party data that ends up in the corpora used to build and refine Grok.

The halt is the latest in a series of European interventions into how generative AI systems are built. Regulators across the bloc keep returning to the same underlying question: whether personal information can be swept into model training without a clear legal basis, and what happens to it once absorbed into model weights, where deletion becomes effectively impossible.

For X, the timing is awkward. Grok is the company's flagship AI product and the main vehicle through which it wants to be seen as an AI company rather than a social network. Any restriction on what may be used to train it narrows the data advantage X has argued it holds through the firehose of user posts.

The scope of the restriction is not settled — whether it covers only EU users or a broader suspension, and whether X could resume third-party training under revised consent or contractual arrangements.

The case fits a wider pattern: European privacy enforcers increasingly treat training-data sourcing itself as the regulated act, separate from how the finished model is later used.