AI privacy6 min read•
AI file scanning is changing cloud privacy
Server-side AI features are powerful, useful, and incompatible with strict confidentiality. Here is what that shift means for sensitive files.
Try it in one click.
Three private surfaces. Same zero-knowledge architecture.
What AI features actually need Smart search, summarisation, classification, image recognition and "ask your files" features all require one thing: server-side access to the plaintext content of your files. That is a fundamental requirement of the current generation of cloud-AI features.
Why this is a privacy shift, not just a feature Encryption at rest only protects files from a stolen hard drive. It does not protect them from features that operate on plaintext inside the provider's environment. As clouds add more AI features, more of your content is processed in plaintext by more systems.
The honest trade-off You can have rich server-side AI features, or you can have a cloud that cannot read your files. With today's models and architectures, you cannot have both at full strength for the same file.
What "private AI" looks like - On-device models that run on already-decrypted content. - Explicit, opt-in features rather than always-on scanning. - Clear separation between content the AI can see and content it cannot.
How DRIVUNO approaches this DRIVUNO does not run AI on your file contents on our side. If we ever add AI features, they will run on the client, on already-decrypted content, and they will be explicitly opt-in per vault.
What is actually scanned today Photos for CSAM and known imagery hashes, documents for copyright matches, code repositories for secret-detection, and increasingly all of the above for AI training data curation. The scanning runs on plaintext, server-side, automatically, at upload time. You did not opt in — you opted in to the cloud.
Why zero-knowledge ends the conversation There is no plaintext for the scanner to see. Whatever pipelines the provider runs, they see ciphertext. The architectural property is the same regardless of intent: cannot scan what cannot be read.
What this means for false positives The single most common cause of cloud account suspensions is an automated scanner mislabel. With DRIVUNO, that failure mode is removed at the root.
Try it in one click.
Three private surfaces. Same zero-knowledge architecture.