Every privacy fine and order against AI chatbot companies, from the regulators

The Morse teamPublished November 11, 20269 min read

As of November 11, 2026, we count five privacy fines against AI chatbot companies on regulators' own websites, and the largest no longer stands. Italy's €15 million fine on OpenAI was set aside when the Court of Rome upheld OpenAI's appeal in a judgment published March 18, 2026, and the Garante, Italy's data protection authority, has taken its decision down. The other four are €5 million on Replika's maker, Luka (under appeal), €158,000 on Character.AI, and two South Korean fines of 103.3 million won and 3.6 million won. Most of what regulators have done to chatbot companies is issue orders: bans, suspensions, and lists of changes to make by a deadline.

We read every decision and press release cited below on the regulator's own site on November 11, 2026: Italy's Garante, South Korea's Personal Information Protection Commission (PIPC), Ireland's Data Protection Commission (DPC), the UK's Information Commissioner's Office (ICO), Brazil's ANPD, Japan's Personal Information Protection Commission, and the Office of the Privacy Commissioner of Canada. We left out anything we couldn't read from the regulator itself, and cases brought by competition or consumer agencies. Several companies changed course during the investigations, and the regulators say so: DeepSeek stopped sending user input to a cloud partner, and OpenAI retired the early ChatGPT models Canada's regulators found were trained unlawfully.

Which AI companies have been fined for privacy?

Four companies: OpenAI, Luka (Replika), Character Technologies (Character.AI), and Scatter Lab, the Korean company behind the "Iruda" chatbot. Only two regulators did the fining: Italy's Garante and South Korea's PIPC. DeepSeek, Meta, and X have faced bans, suspensions, court action, or open inquiries over their chatbots, but no privacy fine.

Every fine and binding order, dated

Built from each regulator's own decision or release, read November 11, 2026. "Status" says only what a primary source confirms.

The rest were findings, warnings, and recommendations, which carry no penalty:

Was OpenAI's €15 million GDPR fine overturned?

Yes. The Garante's own release on the fine now opens with a note: decision No. 755 of November 2, 2024 "has been temporarily removed from the website" following judgment No. 4153/2026 of the Court of Rome, which "upheld the appeal lodged against the Authority's decision" (Garante). The December 2024 news stories that still top search results for this fine, from the Jakarta Post to SecurityWeek, report it as a penalty and don't mention the ruling.

The court ruled only on who could decide the case. As Diritto.it reports, the court accepted OpenAI's first ground, that the Garante lost its competence once Ireland's DPC recognized OpenAI Ireland as OpenAI's EU establishment on February 15, 2024, and treated that ground as absorbing the rest (Diritto.it). Under the GDPR's one-stop shop, cross-border cases then belong to the lead authority. The Garante had already sent its file to the DPC in December 2024 to pursue any violations that continued after OpenAI's Irish office opened. We cover how this sits next to DeepSeek's case in DeepSeek vs ChatGPT privacy.

South Korea's fine on OpenAI gets misreported too. Headlines in July 2023 said ChatGPT was fined for exposing users' data. The PIPC's own release says its experts found it hard to conclude OpenAI had neglected the protections it could reasonably be expected to have, so it didn't sanction the leak itself. The 3.6 million won was for reporting the leak late (PIPC).

What did the orders require besides money?

Age checks, more than anything else. The same measures come back order after order:

• ChatGPT, April 2023. The Garante gave OpenAI until April 30 to show a privacy notice before sign-up, ask users in Italy to declare they're adults, put an age gate in front of existing users, drop "performance of a contract" as its legal basis for training on users' data, and let non-users object to training and ask for wrong answers about them to be corrected or deleted (Garante).
• Replika, June 2023. An age gate on every sign-up page, and a "cooling-off period" so a minor who is turned away can't simply enter a different birth date, with the first four measures due by July 28, 2023 (Garante).
• Character.AI, July 2026. Working age checks, an effective cooling-off period for blocked minors, and minors' profiles set to private by default (Garante).

The other orders went after where chat data travels and what trains on it. Korea told DeepSeek to destroy the user prompts it had sent to Volcano Engine, which DeepSeek had stopped sending on April 10, 2025 (PIPC). Brazil's compliance plan with Meta required notices by email and in the app, a simpler way to object that works even after training starts, and no training on accounts of people under 18 until the ANPD's final decision (ANPD). In Canada, OpenAI committed to warn signed-out ChatGPT users, before their first prompt, that "chats may be reviewed and used to train models" (OPC).

Why are the fines so small?

Because the regulator that acts first often isn't the one in charge. Italy's €15 million fine on OpenAI fell on exactly that point. OpenAI and X both answer to Ireland's DPC as their lead authority under the GDPR, and as of November 11, 2026, its press releases list no final decision on any AI chatbot. Its largest 2026 decision so far is a €403 million fine on Google over location data (DPC).

Some regulators can't fine at all. Canada's Privacy Commissioner says that under the current law he "is not empowered to issue orders," and he has asked for the power to impose penalties (OPC). In Korea, the PIPC handles many AI cases through inspections that end in recommendations, and a company that accepts and carries them out faces no further action.

Which investigations are still open?

Four that could end in fines:

• The DPC on Grok. An inquiry opened February 17, 2026 into sexualized images of real people, children included, made with Grok on X. It tests Articles 5, 6, 25, and 35 of the GDPR (DPC).
• The ICO on Grok. Formal investigations into X and xAI, opened February 3, 2026, into whether "appropriate safeguards were built into Grok's design" (ICO).
• The Garante on Replika's training. The inquiry opened with the 2025 fine asks how Luka assessed risks across the model's development and training, and whether it anonymized the data (Garante).
• The DPC on OpenAI, for any violations that continued after OpenAI's Irish office opened.

In the United States, the FTC has studied companion chatbots but hasn't fined one, as our FTC AI chatbot investigation timeline shows. The state attorneys general suing OpenAI are covered in why states are suing OpenAI.

How Morse keeps your chats from piling up on a server

The Garante's €5 million fine on Luka was for processing Replika users' data without a legal basis, and Korea's order told DeepSeek to destroy the prompts it had sent to a cloud partner. Both went after user data the company had collected and kept.

With Morse, your chat history, including character chats, along with the characters you make and your personas, is stored only in your browser, encrypted, and our servers don't keep your messages or log what you write. Morse doesn't train AI on your content, sell your data, or show ads, so there's no training opt-out to hunt for. Morse is for adults only, and users confirm they're 18 or older.

• Private, for chats you don't want any AI company to keep. New chats start here. Open-weight models like DeepSeek, Kimi, and GLM, plus Grok, run under zero data retention agreements Morse has with its inference, search, and other providers, so they keep nothing. Where a provider confirms zero retention on each reply, Morse checks it and refuses a reply that doesn't.
• E2EE, for anything no one should read. Your browser checks the enclave's hardware attestation before anything leaves your device, then encrypts the message so only that enclave can read it. Our servers relay ciphertext they can't read. The enclaves are run by third-party providers, not by Morse, and the guarantee rests on the enclave hardware and attestation working as designed.
• Anonymous, for GPT, Claude, or Gemini without the provider knowing it's you. Our proxy removes your account, email, and IP address and sends the message under Morse's own account. The provider still sees the text and may keep it for a limited time to check for abuse, under its own policy, so leave out names you'd rather not share.

Because we don't store chats, we usually have very little to give authorities. Clearing your browser or deleting your account deletes your chats, and Morse can't recover them, so keep anything you need. Read the details in our privacy policy.

You can send a first message from the Morse home page without an account, or start free with no card. E2EE chats come with Pro, at $19 a month.