Security, without the flattery

This page summarises our threat model in plain language — including the parts we'd rather not lead with. A privacy product that overstates its guarantees is worse than one that states modest guarantees accurately.

What we protect

The consequence: an attacker who steals our object storage gets encrypted chunks under meaningless names. An attacker who steals our metadata database additionally gets folder-tree shape and wrapped keys — still no content, and no names.

What we can't protect

The browser is the weakest way to use Drive. The web app's JavaScript is delivered by our server on every visit. A malicious or coerced server could serve modified JavaScript that steals your passphrase — and then the encryption protects nothing. This is an unavoidable limitation of end-to-end encryption delivered through a browser, and it applies to every competitor in this space.

The desktop app is the trust anchor. Its code is a build you install once and verify, not a page that can change silently between visits. If the stakes are high, use the desktop app. Subresource integrity and reproducible, verifiable builds for the web app are on the roadmap; today, desktop is the answer.

Under legal compulsion

We can produce

  • Account email address
  • IP addresses and access timestamps
  • File sizes and folder-structure shape
  • The sharing graph between accounts
  • Ciphertext

We cannot produce

  • File content
  • Filenames

At any price, under any order. Not a policy — an architectural fact. We hold no keys.

The short version

Drive makes it impossible for us — or anyone who compromises or compels us — to read your files. It does not hide that you use the service, how much you store, or when. Those are real, accepted limits, and we'd rather you know them than discover them.