When Italy Pulled the Plug on ChatGPT

On 31 March 2023, Italy's data-protection authority — the Garante — ordered OpenAI to stop processing the personal data of Italian users, effectively taking ChatGPT offline in the country. It was the first time a Western regulator had blocked the service, and it reframed AI privacy from a talking point into a concrete compliance event that companies had to act on that week.

The Garante's order, summarised in analyses by firms such as Clifford Chance, rested on several GDPR concerns rather than a single breach. The timing was not incidental: it landed days after the Redis caching bug had exposed some users' chat titles and partial payment data, which the regulator cited as evidence of data-handling risk.

The regulator's objections

• No clear legal basis. The Garante questioned OpenAI's justification, under Article 6 of the GDPR, for the mass collection and processing of personal data to train its models.
• Transparency failures. Users were not adequately informed about how their data was collected and used.
• Accuracy problems. The regulator noted that ChatGPT could produce inaccurate information about real, identifiable people — a data-quality obligation under the GDPR, not just a product quirk.
• Weak age controls. There was no effective mechanism to verify users' ages, despite an intended minimum age of 13.

Resolution, and the bill that followed

The block was temporary. OpenAI worked with the Garante and, after implementing measures — clearer privacy disclosures, a way for users and non-users to object to data processing, and an age-gate at sign-up — ChatGPT was restored in Italy at the end of April 2023. The story did not end there: in December 2024 the Garante concluded its investigation and fined OpenAI €15 million over the underlying data-protection failings. The 2023 ban was the warning; the fine was the invoice.

What defenders and their legal teams should take away

For security and privacy teams, the Garante action set a template that has echoed across every subsequent AI deployment decision. The obligations it enforced are not exotic — they are the ordinary duties of any data controller, applied to a system whose data flows are unusually opaque.

• Know your legal basis before you deploy. If you cannot articulate why you may lawfully process the personal data flowing into an AI system, you have a problem a regulator can act on.
• Map the data. You must be able to say what personal data enters the model, where it goes, and how a subject can object or request deletion.
• Assume the same rules apply to shadow AI. An unsanctioned tool ingesting customer data carries the identical regulatory exposure, with none of the controls a governed deployment would have.

The Garante did not accuse OpenAI of malice; it asked a question every organisation now has to answer for its own AI use. The uncomfortable part is that many cannot — not because they are careless, but because they do not know which AI tools their teams have connected to what data. You cannot demonstrate a lawful basis for processing you have not discovered is happening.