Does the EU AI Act protect your privacy? What it does for your AI chats
The Morse teamPublished 10 min read
As of November 2026, the EU AI Act does little to protect your privacy in AI chats. It makes a chatbot tell you it's an AI, bans a short list of manipulative and biometric practices, and makes the companies behind large models publish a summary of what they trained on. It doesn't decide how long a chatbot keeps your chats, whether they train a model, or whether you can see or delete them. The Act itself says it "shall not affect" the GDPR, and the GDPR is the law that covers all three (AI Act, Article 2(7)).
We read the AI Act, the Digital Omnibus on AI that amended it, and the GDPR on November 18, 2026. We read archived copies of each text (captured November 12 and 16, 2026, and January 1, 2026 for the GDPR). We also read the European Commission's AI Act pages, its implementation timeline, the General-Purpose AI Code of Practice, the Commission's guidelines on Article 50, its template for training-data summaries, and the European Parliament's file on the separate GDPR omnibus. The Commission deserves credit here. Its timeline page marks every date the Omnibus moved, and its Article 50 guidelines give concrete examples of what counts as an obvious chatbot. This isn't legal advice.
The AI Act's key dates, and what each means for your chats
Every row comes from the Act's text, the Omnibus, or the Commission's AI Act timeline, read November 18, 2026.
| Obligation | Who it applies to | Applies from | What it means for your chats |
|---|---|---|---|
| Banned practices: manipulation, exploiting vulnerabilities, social scoring, emotion recognition at work and school, inferring sensitive traits from biometrics (Article 5) | Anyone placing such a system on the market or using it | February 2, 2025 | A chatbot can't be built to push you into a harmful decision; nothing about storing your messages |
| AI literacy (Article 4, softened by the Omnibus) | Providers and deployers, for their own staff | February 2, 2025 | Your employer may train you on AI tools; nothing about your data |
| General-purpose model duties: documentation, a copyright policy, and a public training summary (Article 53) | Companies that make models like GPT, Claude, and Gemini | August 2, 2025; Commission enforcement from August 2, 2026 | A public yes or no on whether user prompts trained the model, and no opt-out |
| Chatbot disclosure (Article 50(1)) | Providers of AI systems that talk to people | August 2, 2026 | You're told it's an AI at the first interaction, unless that's obvious |
| Machine-readable marking of AI output (Article 50(2)) | Providers of systems that generate text, images, audio, or video | August 2, 2026; December 2, 2026 for systems already on sale (Omnibus) | Replies can carry a machine-readable mark that says they're AI-made |
| Deepfake labels (Article 50(4)) | Businesses and others using AI to make deepfakes | August 2, 2026 | A fake image, video, or voice of a real person must say it's fake |
| Ban on AI that makes non-consensual intimate images or child sexual abuse material (Omnibus) | Providers and deployers | December 2, 2026 | Apps made to undress real people can't be sold or used for that in the EU |
| High-risk rules for hiring, credit, education, and other Annex III uses (Omnibus) | Providers and deployers of those systems | December 2, 2027 (was August 2, 2026) | Logs kept at least six months, and a right to an explanation of a decision |
| High-risk rules for AI built into regulated products (Omnibus) | Product makers | August 2, 2028 | Mostly not about chatbots |
Two of those changes arrive in two weeks. On December 2, 2026, the nudification ban takes effect and the grace period for marking AI output ends.
What is the EU AI Act?
The AI Act is Regulation (EU) 2024/1689, the EU's law on artificial intelligence, which entered into force on August 1, 2024 (European Commission). The Commission describes four levels of risk, from banned practices down to "minimal or no risk," for which "the AI Act does not introduce rules." Most of the Act became applicable on August 2, 2026. Its duties fall on the companies that build, sell, and deploy AI. Article 2(1) covers providers "irrespective of whether those providers are established or located within the Union or in a third country," and providers abroad whose system's output is used in the EU (AI Act). A person chatting with an AI for personal reasons has no obligations under it, because the Act doesn't apply to people using AI "in the course of a purely personal non-professional activity" (Article 2(10)).
Does the AI Act cover what happens to your chats?
Mostly not. Article 2(7) says EU law on personal data "applies to personal data processed in connection with the rights and obligations laid down in this Regulation," and that the Act "shall not affect" the GDPR or the ePrivacy Directive (AI Act). Guides still file the Act among the laws that guard your data. A 2025 feature from INTA, the trademark association, says it "reaffirms the fundamental right to the protection of personal data" (INTA). Read article by article, the Act sets no limit on how long a chatbot keeps your messages, no rule against training on them, and no right to see or delete them.
How long can a chatbot keep your chats under the AI Act?
The Act doesn't say. Where it does mention keeping records, it sets a minimum. Providers and deployers of high-risk systems must keep the logs their systems generate "of at least six months," unless data protection law says otherwise (Articles 19 and 26(6)). A general chatbot isn't high-risk, so even that rule doesn't reach it.
Can a chatbot train on your chats under the AI Act?
Yes, as far as the AI Act goes, but it has to say so. Article 53(1)(d) makes providers of general-purpose AI models publish "a sufficiently detailed summary about the content used for training," using a template from the Commission's AI Office (AI Act). That template asks, with a Yes and a No box: "Was data from user interactions with the AI model (e.g. user input and prompts) used to train the model?" (European Commission). Ticking Yes breaks no rule in the Act. Your right to object to that training comes from Article 21 of the GDPR, and we walk through each company's form in how to object to AI training under the GDPR.
Does the AI Act give you any rights at all?
Two narrow ones. Article 86 gives a right to "clear and meaningful explanations" when a deployer makes a decision about you based on a high-risk system's output, such as one that screens job applications (AI Act). The high-risk rules those systems answer to now start on December 2, 2027 (Omnibus). Article 85 lets anyone complain to a national market surveillance authority about a breach of the Act. Neither one gets you a copy of your chats.
What does the AI Act require of chatbots?
It has to tell you it's an AI. Article 50(1) says providers must design systems "intended to interact directly with natural persons" so that people "are informed that they are interacting with an AI system," unless that's obvious to someone "reasonably well-informed, observant and circumspect" (AI Act). The notice must come "at the latest at the time of the first interaction" (Article 50(5)). The rule has applied since August 2, 2026, and the Commission sums it up as "humans should be made aware that they are interacting with a machine" (European Commission).
The Commission's guidelines name general-purpose chatbots as within the rule and narrow the "obvious" exception to cases "where there is almost no doubt left" (Article 50 guidelines). Their examples of obvious use are a code assistant for professional developers and a non-player character in a single-player game. A chatbot on a helpdesk, whose replies a user "may perceive as human-generated," gets no exception.
The marking rule in Article 50(2) has a privacy side. The same guidelines say marks must follow data protection law, and that "the information related to the creator of the content should not be processed." The mark is meant to show that a machine wrote the text, without recording who asked for it. We look at what one real watermark carries in Claude's text watermark. Breaking any Article 50 duty can cost a company up to €15 million or 3% of worldwide turnover, whichever is higher (Article 99(4)).
Does the EU AI Act apply to ChatGPT, Claude, and Gemini?
Yes, both as general-purpose models and as chatbots. The Commission's guidelines say Article 50 covers general-purpose AI systems "such as chatbots supporting direct interaction with natural persons" (Article 50 guidelines). The models behind them carry the Article 53 duties, and a model trained with more than 10^25 floating point operations is presumed to carry "systemic risk" (Article 51(2)). Those providers must also run adversarial testing, report serious incidents, and protect the model's cybersecurity (Article 55). The Commission, rather than a national authority, can fine a model provider up to €15 million or 3% of worldwide turnover (Article 101).
The Commission published a voluntary General-Purpose AI Code of Practice on July 10, 2025, with three chapters: transparency, copyright, and safety and security (European Commission). Its signatory list, read November 18, 2026, includes Amazon, Anthropic, Google, Microsoft, Mistral AI, and OpenAI. Meta isn't on it, and xAI signed only the safety and security chapter. The Commission's page on the Code doesn't mention personal data or privacy.
What did the Digital Omnibus change?
It moved deadlines and added a ban, but it left chat privacy where it was. The Digital Omnibus on AI is Regulation (EU) 2026/1744 of July 8, 2026, published in the Official Journal on July 24 (EUR-Lex). It entered into force on July 27, 2026 (European Commission). It pushed the high-risk rules to December 2, 2027 and August 2, 2028, gave existing generative systems until December 2, 2026 to mark their output, and banned AI that makes intimate images of an identifiable person without their "freely-given, specific, informed, unambiguous and explicit consent." It also added an Article 4a that lets high-risk providers "exceptionally process" special categories of personal data, such as health or ethnicity, to detect and correct bias, under conditions.
The other Digital Omnibus, the one that would amend the GDPR, hasn't passed. As of November 18, 2026, the European Parliament's procedure file lists it as "Awaiting committee decision." Your GDPR rights below are the ones in force today.
What protects your chat data in the EU, then?
The GDPR. Where the AI Act is silent on each question a chatbot user asks, the GDPR has an article. We read both on November 18, 2026.
| What you want to know | What the AI Act says | What the GDPR says |
|---|---|---|
| How long can they keep my chats? | Nothing for chatbots; high-risk logs kept "at least six months" | Data kept "no longer than is necessary" (Article 5(1)(e)), and you're told "the period for which the personal data will be stored" (Article 13(2)(a)) |
| Can they train on my chats? | Must tick Yes or No in a public training summary | Right to object to processing based on legitimate interest (Article 21) |
| Can I get a copy? | No | Right of access (Article 15), answered "within one month of receipt" (Article 12(3)) |
| Can I make them delete it? | No | Right to erasure "without undue delay" (Article 17) |
| Who do I complain to? | Market surveillance authority (Article 85) | Data protection authority (Article 77) |
| Top fine | €35 million or 7%, whichever is higher, for banned practices (Article 99(3)) | €20 million or 4% (Article 83(5)) |
To use the access right on each major chatbot, see how to download your AI chat data. The privacy fines regulators have issued against chatbot companies came under data protection law, and we list each one in every privacy fine against AI chatbot companies. For the US side, where state laws do the work, see AI chatbot laws by state.
How Morse answers the questions the AI Act leaves open
Each gap above exists because your conversation sits on a company's servers, where the AI Act asks only that the company disclose what it did with it. The GDPR then gives you ways to ask for it back, object, or have it deleted, one request at a time. Morse is built so your chats don't pile up on our servers.
On retention, Morse's servers don't keep your messages or log what you write. Your chat history is stored in your browser, encrypted, not on Morse's servers. On training, Morse doesn't train AI on your chats, sell your data, or show ads. On deletion, clearing your browser or losing the device deletes your chats for good, and Morse can't recover them, so keep a copy of anything you need. Because we don't store chats, we usually have very little to give authorities.
- Private, for questions you'd rather no model host keep. Every new chat opens in this mode, with open-weight models such as DeepSeek, Kimi, and GLM, plus Grok. Morse has zero data retention agreements with its inference, search, and other providers, so they've agreed to keep nothing. Where a provider confirms zero retention on each reply, Morse checks it and refuses a reply that comes without it.
- E2EE, for the conversation you'd show no one. Your browser checks the enclave's hardware attestation, then encrypts the message so only that enclave can read it, and Morse's servers relay text they can't read. The enclaves are run by third-party providers, not by Morse, and the protection rests on the enclave hardware and attestation working as designed. E2EE runs open-weight models only, with no web search or image generation.
- Anonymous, for GPT, Claude, or Gemini without your name attached. Our proxy removes your account, email, and IP address and sends the message under Morse's own account, so the provider sees it without knowing it's you. The provider may keep it for a limited time to check for abuse, under its own policy, so leave names and identifying details out of the message.
Send a first message from the Morse home page without an account, or start free. End-to-end encrypted chats come with Pro, at $19 a month.
Common questions
Does the AI Act put any duties on me as a chatbot user?
Not for personal use. Article 2(10) says the Act doesn't apply to the obligations of deployers who are people using AI 'in the course of a purely personal non-professional activity.' At work it can, because your employer, as a deployer, has the AI literacy duty in Article 4.
Does the AI Act apply to AI companies outside the EU?
Yes. Article 2(1) covers providers that place AI systems or general-purpose models on the EU market 'irrespective of whether those providers are established or located within the Union or in a third country,' and providers abroad whose system's output is used in the EU.
Where do you complain about a chatbot in the EU?
It depends on the problem. Under Article 85 of the AI Act, a complaint about a breach of the Act goes to the national market surveillance authority. A complaint about your data, such as an unanswered access or deletion request, goes to a data protection authority under Article 77 of the GDPR.
What are the AI Act's fines?
Up to €35 million or 7% of worldwide annual turnover, whichever is higher, for a banned practice, and up to €15 million or 3% for most other breaches, including the chatbot disclosure rules in Article 50. The Commission itself can fine providers of general-purpose AI models up to €15 million or 3%.