Mistral and Mozilla announced on September 16 that Mistral models are powering Firefox Smart Window, Mozilla's AI browsing assistant in beta. Their joint announcement names France and North America first, with the United Kingdom and Germany expected later in 2026.
The assistant can help with searches, finding previously visited material and questions informed by browser tabs, the companies say. They also describe a privacy arrangement: Mozilla does not save conversations on its servers by default, while partners such as Mistral commit to zero data retention.
A retention promise is not an on-device claim
Those terms matter, but they should be read precisely. The announcement does not establish that every request runs locally or that no data leaves the browser. Nor have we independently audited the implementation. A statement about how long a provider keeps information answers a different question from what information is transmitted.
Our assessment is that the interesting product change is where the assistant gets its context. A browser can put an answer beside the material a person is already reading. That could reduce copying between tabs and a separate chatbot, but it also makes the boundary around selected content worth understanding.
What to check in the beta
An illustrative first test would use public pages about an unfamiliar topic. Ask a question spanning two tabs, follow the answer back to its sources and check whether it preserves the differences between them. Then inspect the controls for choosing which context the assistant receives. This is a proposed evaluation, not a hands-on result from our newsroom.
For a reader choosing an AI browser, a useful comparison should cover three separate things: answer accuracy, control over context, and the stated handling of submitted data. A strong result in one category does not settle the others.
The rollout is regional and still in beta. The next useful evidence will be how clearly those controls work in practice, alongside the expansion the companies have announced. Model choice is valuable when users can understand what they are choosing.
