Shareboxed
Cloud services have quietly grown a second customer: their own AI features. Files get scanned to power search, summaries, photo recognition and assistants, and the terms of service grow a paragraph explaining why that is fine. Sometimes you can opt out. You are rarely asked to opt in.
Shareboxed takes the position that a file you share is between you and the person you share it with, and we built the product so that the position does not depend on our good behaviour.
"We do not" versus "we cannot"
Any service can write "we do not run AI on your content" in a policy. The sentence is only as durable as the next business decision. What holds up is architecture: files in a hosted share are encrypted on your Mac before upload, and the key travels in the link fragment that never reaches our servers. There is no readable copy on our side to analyse, summarise, index or train on. A direct share goes further still: the files never touch our infrastructure at all.
We are precise about the edges, because vague privacy claims are how trust erodes. For a hosted share we can see file names and sizes; for a share out of your own bucket we can read the folder prefix; and we say both in the privacy policy. What we cannot do, structurally, is open your files.
No AI features either
There is also simply no AI in the product. No content scanning, no automatic tagging, no assistant reading along. Sharing a file is a small, mechanical job, and we think the tool that does it should be equally small and mechanical: encrypt, upload, hand back a link, delete on expiry.
If a future ever arrives where an AI feature would genuinely help, it would have to work the way everything else here works: on your device, with your keys, and off by default. That is the standard the architecture enforces, and it is the reason this page can make promises a settings toggle cannot.