Elon Musk’s X will voluntarily drop EU data from its “Grokbot” AI training regimen, but only after the Irish Data Protection Commission (DPC) took the highly unusual step of requesting a court order from the national High Court.
Known more for being generous with the American tech firms that take up European regional residence in and around Dublin, the Irish DPC’s request to the court for an order to prohibit X’s processing of personal data was the first of its kind. Musk has been at odds with EU regulators since he took over Twitter, first for refusal to police speech on the platform to its desired regulatory standards and now for processing of user data in what may be violation of the General Data Protection Regulation (GDPR).
Grokbot AI training collected European posts over the summer
Grokbot reportedly collected posts from European users for its AI training from May 7 to August 1 this year. This was done without any notification of users or collection of consent, creating a potential GDPR violation and sending the usually relaxed Irish DPC to the country’s High Court for an urgent order to suspend its processing of personal data entirely.
The move by X seems to have done some immediate damage control, but the DPC and other European regulators are proceeding with investigations into the platform’s use of European user data. The DPC said that the matter at hand was now “adjourned,” however, and would be picked up again by the courts in September. While X appears to have agreed to ceasing AI training in front of a judge, the company has also protested the decision in a public statement. It casts itself as more respectful of EU data and general user privacy as it uses only internal platform data to train its chatbot, and allows users to opt out of having their posts included (via a privacy setting that is on by default). However, the option to opt out did not appear until July 16, or over two months from the beginning of EU data collection, and was also not rolled out to all platform users immediately.
Grokbot is currently only available to paid premium subscribers of the platform, as its AI training continues. X pitches it as a humorous assistant that does not necessarily have the same speech restrictions as other chatbots. Given that it only trains on the platform’s posts, critics have raised concerns that the bot will struggle more than others with discerning what is and is not factual, and may be more prone to propagate hateful or harmful information. Musk has advertised a forthcoming new version of Grokbot that significantly improves its ability to code and solve math problems, drawing it closer to the capability of leading competitors such as ChatGPT.
AI processing revives old EU data debate
The legal debate over “lawful purpose” exceptions to consent for processing EU data seemed to have wound down after Meta essentially tried out every possible option, having them all eventually rejected by either regulators or courts. But as X demonstrates, the relatively new and untested issue of AI training may open up roads for these sorts of challenges again.
X’s Grokbot is far from alone in having its AI training methods questioned. ChatGPT was first into the fray and attracted enforcement attention from numerous European countries, including a temporary ban in Italy over concerns of GDPR consent violations and inappropriate handling of the personal information of minors. The European Data Protection Board (EDPB) was eventually involved, but has yet to answer numerous key questions about the fine details of EU data collection and AI training.
That regulatory uncertainty has caused other tech companies to pause operations in the EU. Meta has indefinitely paused the launch of its AI models in Europe, citing the still-developing state of regulation of EU data it will inevitably take in, and discussions with the Irish DPC have also led Google to temporarily remove Gemini from the EU market as it makes changes to address potential issues.
Ireland’s Data Protection Act 2018 allows the country to individually restrict or stop data processing if a case is deemed “urgent” to protect people’s rights, but the tech-friendly haven has never actually used this ability until now. Prior to the development of the AI training issues, Musk had already established a rocky relationship with the EU by refusing its requirements to limit what it calls disinformation and hate speech on the platform. Musk’s seeming reticence to play ball with the full scope of the EU’s terms triggered a European Commission probe launched just ahead of Christmas last year, which is investigating not only the speech issues but also its handling of paid “blue check” verifications. The platform is being investigated under the Digital Services Act (DSA) which allows for fines of a maximum of 6% of a company’s annual turnover.
