X caught feeding user posts to Grok by default

If you have posted on X recently, there is a very good chance your words are already in the stomach of an AI model without your knowledge.

X users discovered that a new privacy setting had been turned on by default. The pre-ticked box quietly granted Elon Musk’s xAI permission to use public posts, interactions, prompts, and results to train and fine-tune its Grok chatbot. To make matters worse, users reported that the toggle could only be found and disabled on the web version, leaving mobile app users largely in the dark.

Regulators caught off guard

European data watchdogs did not take long to react. The UK Information Commissioner’s Office (ICO) and Ireland’s Data Protection Commission (DPC) both launched inquiries into X over the stealthy move.

Under UK and EU GDPR rules, obtaining consent through pre-ticked boxes or any default opt-in mechanism is strictly prohibited. The ICO stated that platforms training AI models must proactively notify people well in advance and provide straightforward opt-out processes.

Ireland's DPC, which serves as the lead EU regulator for X, sounded particularly blindsided. Deputy Commissioner Graham Doyle revealed that the agency had been talking with X about AI data collection for months—including a meeting just the day before the discovery—making this sudden default rollout a genuine surprise.

Why it matters

Data hunger is pushing frontier AI labs into dangerous territory. With copyright lawsuits piling up from publishers and authors, user-generated platform data has become gold dust. Just earlier this month, Meta cited unpredictable EU regulations when deciding not to ship its advanced AI models in the region. X apparently took the opposite approach: take the data first, deal with regulators later.

In our view, this was a reckless gamble by Musk's team. X has a legitimate goldmine of real-time human conversation that gives xAI a unique edge against OpenAI and Anthropic. But trying to sneak user consent past European regulators via a hidden web toggle is legally indefensible under GDPR and invites massive fines.

For everyday users and creators, this is an immediate call to action. Head to your account privacy settings on desktop and uncheck the Grok training permission right now. It is a stark reminder that in the generative AI race, your public thoughts are treated as free training fuel until you explicitly fight back.