What’s another AI chatbot in an already crowded field? That’s the question Proton is trying to answer today with the release of its new . And like with its best known service, , the company says Lumo is for those who want a private alternative to what big tech is offering.

Proton says every conversation with Lumo is secured with zero-access encryption, meaning only your device can unlock your content. In the context of an AI chatbot, that has several implications. Most notably, it means not even Proton can view your chats. As a result, the company can’t share your data with governments, advertisers or, for that matter, any other company, and it can’t use your data to train future AI models. “By using Lumo, you can enjoy the benefits of an advanced AI assistant without the risk of your data being misused,” says Proton.

I briefly tried Lumo. It’s a bit slow to generate a response, but you can broadly expect a similar experience to what you would find using ChatGPT or Claude for free. Lumo can search the web to answer questions beyond its knowledge cut-off date, but by default that feature is turned off to further protect user privacy. You can also upload files to Lumo. Here again Proton says the chatbot won’t save any information.

Proton isn’t touting the performance of Lumo’s large language models, but if you’re curious about this sort of thing, it’s powered by a handful of open-source systems, including Mistral NeMo and Mistral Small 3, among others. Proton  Lumo will filter requests through the model best suited for the task. For example, it will use NVIDIA’s OpenHands system for coding requests.

Lumo is free to use, with a weekly query limit. You don’t need a Proton account to begin a conversation with the chatbot. In addition to being , Proton offers both Android and iOS apps. A $13 per month Plus plan offers unlimited usage, alongside perks like larger uploads, access to more advanced AI models, priority support and more.

If you buy something through a link in this article, we may earn commission.

Artificial intelligence (AI) based on large language models (LLMs) has revolutionized the industry. Of these, ChatGPT is probably the best-known and arguably the most sophisticated. Today people turn to ChatGPT to answer questions, have conversations, and share deeply personal stories.

But can you trust ChatGPT with your secrets? A creator recently posed this question in a viral TikTok.

“Imagine telling ChatGPT all your business,” Ms. V (@sweetlifeofv) says in the post. “And somebody else somewhere asking ChatGPT for book ideas and they tell them your story.”

She captions it, “Imagine picking up a book in a bookstore and being like damn this sounds familiar.”

The consensus among top comments on the post is clear: ChatGPT can’t be trusted.

Does AI kiss and tell?

ChatGPT creator OpenAI is currently being sued for copyright infringement. Several newspaper companies, including the New York Times, claim OpenAI used their work to train ChatGPT. They say it has at times provided their work verbatim in response to users’ questions. This suit was consolidated with cases brought by authors including John Grisham, Jodi Picoult, Jonathan Franzen, and Sarah Silverman.

OpenAI has denied infringing the companies’ and authors’ copyrights. The company told the Associated Press, “We build our AI models using publicly available data, in a manner grounded in fair use, and supportive of innovation.”

Researchers have found that ChatGPT does disclose personal information. A 2023 paper by a team from Google claims they were able to obtain the personal information of thousands.

“Using only $200 USD worth of queries to ChatGPT (gpt-3.5- turbo), we are able to extract over 10,000 unique verbatim memorized training examples,” they wrote.

On Saturday, the Mary Sue asked ChatGPT for a book idea. It suggested a tale of a librarian who discovers a book with no title. She initially assumes it’s fiction—until pages start appearing chronicling events in her own life.

ChatGPT further offered to tailor the idea to the Mary Sue’s preferences by making it science fiction or more “grounded.”

OpenAI did not respond to an emailed inquiry sent Saturday morning.

The AI remembers

Whether ChatGPT and other LLM-based AI remember conversations is a matter of some debate. One person in the comments on Ms. V’s TikTok claims they work in AI and it does not.

Its terms say otherwise.

ChatGPT is trained with users’ conversations, according to OpenAI’s website. You can opt-out, however. The company instructs users to click “do not train on my content” in its privacy portal.

You can also ask it to delete their conversations, but at present it will not comply. The judge in the New York Times case recently ordered OpenAI to retain all conversations indefinitely. OpenAI is appealing.

ChatGPT’s terms prohibit users from falsely claiming an idea “was human generated when it was not.” But in the case of a book or other work of art, it could be difficult to prove that the author essentially stole your story from AI.

Ms. V’s post has 3 million views as of this writing. Nearly 4,000 people have commented.

Reactions range from horror to unsurprised.

“You do not have to imagine it, it’s definitely happening,” a comment with 90,000 likes reads. “Your conversations are NOT private.”

A second agreed, “This is 100% what’s happening.”

Another likened it to an episode of Black Mirror.

Many people brought jokes.

“No because after I’m done I told ChatGPT, ‘This is between you and me only ok? Do not tell anyone,’ and it promised me lol,” wrote one.

Ms. V replied, “Not ChatGPT pinky promised!”

Another said, “Please don’t make me think my friend is a snitch. ChatGPT is my bff.”

“I just know it be a psychological thriller,” a third quipped.

@sweetlifeofv Imagine picking up a book in a bookstore and being like damn this sounds familiar ?? #chatgpt #bookideas #booktok ♬ original sound – sweetlifeofv

Ms. V didn’t immediately respond to an email sent Saturday morning.

Have a tip we should know? [email protected]