01What SafeSend receives
SafeSend only ever sees content you hand it deliberately. It does not read channel history, and it does not monitor anybody else's messages.
When you run /safe-send, SafeSend receives:
- the draft text you type into the command
- up to five images you attach to it
When you run /privacy-settings, SafeSend receives your choice of privacy categories and the sensitivity level for each.
In both cases Discord also tells SafeSend which user invoked the command.
02Third-party services
To find faces and to read text that is visible inside a picture, SafeSend sends the images you attach to Google Cloud Vision. Google states that content sent to the Cloud Vision API is not used to train or improve its models.
Your draft text is not sent anywhere. It is checked on our own server using local pattern rules, and it never leaves that server except as the message you choose to post.
No other third party receives your content. SafeSend does not sell data, does not share it with advertisers or data brokers, and does not use anything you submit to train any model.
A separate research build of SafeSend, used only inside a supervised university study, additionally sends draft text to a language model for classification. Participants in that study are told about this separately before taking part. That build is not the public SafeSend and is not what you get when you add the app from Discord.
03What is stored, and for how long
| What | Where | Kept for |
|---|---|---|
| Your draft text and the snippets detected in it | Session record on the server | Deleted automatically within 24 hours |
| Images you attached, the protected copies, and review previews | Session files on the server | Deleted automatically within 24 hours |
| Your privacy category settings | Settings file on the server | Until you change or reset them |
| Research records (see section 4) | Database on the server | Indefinitely, unless you ask for removal |
A submitted draft only needs to survive long enough for you to review it, so its session files are deleted automatically within 24 hours.
Your Discord account ID is never stored in its raw form. It is converted with a one-way keyed hash into a code that cannot be reversed back into your account, and only that code is stored. SafeSend does not store your Discord username, your display name, or the identifiers of any server or channel.
04Research records
SafeSend records a small amount of data about how its warnings are used. Each record contains only:
- the one-way code described above
- a session identifier
- for each detected item: whether it came from text or an image, which category it fell into, whether it triggered a warning, and which action you chose
These records contain no message content: no draft text, no images, no text recognised inside images, and no coordinates.
05Who can see the warnings
Every warning, settings panel and review screen is sent as a Discord ephemeral message, which means only the person who ran the command can see it. Nothing appears in the channel until you choose to post it.
When you do choose to post, the message is sent by SafeSend on your behalf and is labelled with your display name, your avatar and the words "via SafeSend", so that the channel knows who wrote it.
06Deleting your data
To delete your settings, run /privacy-settings and choose Reset to Default.
To have your research records removed, or to ask what is held about you, contact us at the address in section 8. Because records are stored under a one-way code, please send the request from the Discord account concerned, or include your Discord account ID so the corresponding code can be derived.
Session files delete themselves within 24 hours and need no request.
07Children
SafeSend follows Discord's Terms of Service, which require users to meet the minimum age for their country. It is not directed at children.
08Contact
Questions, access requests and deletion requests: luoxingli610@gmail.com
09Changes
If this policy changes, the date at the top will be updated. Material changes to what SafeSend collects or who it is shared with will be announced in the support server before they take effect.
Read the Terms of Service