Why AI features and zero-knowledge are in tension
Most AI features on a cloud — smart search, summarisation, classification, image recognition — require server-side access to your content. That is fundamentally incompatible with a strict zero-knowledge model.
The trade-off
You can have AI on your cloud, or you can have a cloud that cannot read your data. You cannot have both at full strength, today, with the current generation of models.
Where on-device AI fits
On-device models can offer useful features (search, summarisation, suggestions) without breaking zero-knowledge — because the model runs on the client, on already-decrypted content. This is the direction privacy-respecting AI is moving.
DRIVUNO's position
DRIVUNO does not run any AI on your file contents on our side. If we ever introduce AI features, they will be on-device and explicitly opt-in.
The quiet shift no one consented to
Most large clouds now run automated scanners over user files — for CSAM detection, copyright matching, malware, and increasingly for AI training data preparation. The intent is sometimes legitimate. The capability is the problem: anything they can scan, they can scan for anything else later. The architecture is reusable.
False positives are not theoretical
There are documented cases of users losing access to entire Google or Apple accounts after automated systems mislabelled medical photos of their own children. The appeals process is opaque, slow, and sometimes never resolves. Once an automated system flags you, the human review is a formality.
The zero-knowledge alternative
With DRIVUNO, your files are unreadable to any server-side scanner because no server-side scanner can decrypt them. False positives are not just rare — they are mathematically impossible at the storage layer.
Try it in one click.
Three private surfaces. Same zero-knowledge architecture.