NextFin News - Proton has upgraded Lumo, its privacy-focused AI chatbot, with image recognition, image generation, persistent memory in Projects, and a new thinking mode, making the assistant materially more capable while keeping the company’s core privacy promise intact. The new version, Lumo 2.0, is available now and is meant to show that a confidential AI product does not have to feel stripped down to be credible.
That is the real significance of the release. Proton is no longer positioning Lumo as only a private alternative to mainstream chatbots. It is now trying to compete on usefulness as well, while preserving the zero-access encryption, no server-side logging, and no training-on-customer-data rules that define the product.
Proton says Lumo 2.0 responds to most queries up to 76 percent faster than the previous version. The company also says the upgrade performs better on the Artificial Analysis Intelligence Index, with Lumo 2.0 Lite scoring 127 percent higher than Lumo 1.4 and Lumo 2.0 Max scoring 240 percent higher. In other words, Proton is trying to prove that privacy and performance can move together instead of trading off against each other.
The new capabilities broaden what users can do inside the assistant. Images can now be uploaded for analysis or editing, and Lumo can also generate images from prompts. Projects, which connect Lumo to Proton’s other services, now include user-controlled persistent memory so the assistant can remember preferences across sessions. Proton says the platform continues to run on its European infrastructure and remains protected by Swiss privacy laws and zero-access encryption.
Andy Yen, Proton’s founder and chief executive, said the upgrade reflects both a technical rebuild and a broader product ambition.
“Lumo 2.0 has been re-engineered from the ground up and the introduction of thinking mode gives it powerful new capabilities,” Andy Yen said. “Lumo 2.0 demonstrates that users no longer need to choose between powerful AI capabilities and meaningful privacy protections.”
For a privacy-first product, the move is strategic. The consumer AI market has moved beyond basic question answering. Users now expect image tools, memory, and stronger reasoning, and that shift makes it harder for a privacy-centric assistant to compete on principle alone. Proton’s answer is to add the features people expect while keeping the product’s data model deliberately restrictive.
That matters because the data handled by AI assistants is growing more sensitive, not less. People are uploading files, summarizing documents, comparing images, and revisiting long-running conversations. In that setting, Proton’s claims about zero-access encryption and the absence of server-side logs are not just security talking points; they are the foundation of the product’s value proposition.
Proton says Lumo does not use customer data for AI training and does not share it with third parties. Those statements are likely to resonate with users who want a practical assistant but are wary of systems that depend on broad data collection to improve. The company is clearly betting that trust can become a purchase driver, not just a compliance box.
The release also shows how AI differentiation is changing. When model access alone was the main selling point, privacy-first products could live on brand and ideology. Now the pressure is on utility. A product has to be private, fast, and capable at the same time. Lumo 2.0 is Proton’s attempt to meet that threshold.
A Privacy Product Now Has to Be a Full Product
The biggest challenge for privacy-focused AI is that users no longer judge it only against privacy-sensitive alternatives. They compare it with the best general-purpose assistants they already know. That means features such as image recognition, image generation, and memory are no longer nice-to-have extras. They are table stakes.
Proton’s answer is to fold those capabilities into a system that still claims it never logs conversations on the server side and never uses customer data to train AI models. That is a meaningful distinction. Many users may be willing to give up some convenience for privacy, but only if the product is good enough to use daily. Lumo 2.0 is an attempt to make that compromise look unnecessary.
The company’s benchmark claims are part of that effort. A 127 percent improvement for Lumo 2.0 Lite over Lumo 1.4 and a 240 percent improvement for Lumo 2.0 Max are substantial by any measure, even if the details of the benchmark should be read as Proton’s own performance framing rather than an independent market ranking. The point is less the absolute score than the message: this version is supposed to feel like a different class of product.
That distinction matters for adoption. A chatbot that is clearly slower or less capable than mainstream rivals will struggle no matter how strong its privacy story is. But if the gap narrows enough, privacy becomes part of the purchase decision rather than an excuse to settle for less. Proton appears to understand that dynamic and has organized the upgrade around it.
The Projects feature is especially important because it moves Lumo closer to a workflow tool. Persistent memory lets the assistant recall preferences across sessions, which is useful in long projects and recurring tasks. That turns Lumo from a one-off chat interface into something closer to a private workspace, especially for users already inside Proton’s broader product ecosystem.
Proton’s strategy is also broader than the chatbot itself. By tying Lumo to the company’s email and cloud products, it can use existing trust and familiarity to encourage adoption. A user who already depends on Proton for communications or storage may be more willing to try Lumo than someone starting from scratch with a new AI brand.
Still, the upgrade does not erase the product’s limitations. Proton is not claiming to outperform the biggest AI platforms across every task. Instead, it is trying to establish a usable middle ground: enough capability to be practical, enough privacy to be differentiated, and enough speed to feel responsive.
What Proton Is Really Selling
Proton is ultimately selling a different operating model for AI. In that model, the provider does not need to read user conversations, retain them for training, or build a business around accumulating more and more behavioral data. That is a more ambitious promise than a standard “private mode” toggle because it changes how the service is built.
It also reflects a broader shift in the AI market. As the basic experience of asking questions and getting answers has become commonplace, the meaningful differentiators have moved toward workflow integration, latency, personalization, and data handling. Proton is leaning hardest into the last of those, while trying to close the gap on the rest.
That approach has a clear logic. If Lumo feels fast enough, useful enough, and private enough, Proton can carve out a durable niche among professionals and consumers who are uncomfortable with more data-intensive AI systems. If the experience still feels too limited, privacy may remain an appealing idea without becoming a default choice.
The release suggests Proton is trying to move past that risk. Image tools make the chatbot more practical. Thinking mode makes it more capable on harder prompts. Persistent memory makes it more useful over time. And the benchmark claims are meant to reassure users that the core engine has improved, not just the feature list.
In that sense, Lumo 2.0 is less a cosmetic refresh than a product reset. Proton is trying to tell the market that confidential AI does not have to be a downgraded version of mainstream AI. It can be a category with its own standards, especially for users who treat privacy as a functional requirement rather than a philosophical preference.
The next test is whether users agree. If Lumo 2.0 can be both useful and private, Proton has a clearer case for building AI into the same trust-first ecosystem that made its other products stand out. If not, the upgrade will still mark progress, but it may not be enough to make private AI feel indispensable.
Explore more exclusive insights at nextfin.ai.
