A few years back people were getting messages from Google if their docs contained hate speech.

Google's response was that it was a corporate feature that was accidentally turned on for all accounts.

So they certainly have infrastructure to do deep content scans of all users docs. Realistically they probably do still scan all accounts for internal metrics, it's just notification that has been disabled.

Do you have a source for that?

I can't find anything with searching that Google ever algorithmically identified hate speech in Docs & Drive, there's no such corporate feature that can be enabled in G Suite control panel, reliably identifying hate speech is a hard problem that there's no indication of Google having solved, and honestly the entire thing sounds like an urban legend.

But if you have a reputable source I'd love to know.

> A Google spokesperson reached out via email with the following statement saying that the bug has been fixed: "This morning, we made a code push that incorrectly flagged a small percentage of Google Docs as abusive, which caused those documents to be automatically blocked. A fix is in place and all users should have full access to their docs. Protecting users from viruses, malware, and other abusive content is central to user safety. We apologize for the disruption and will put processes in place to prevent this from happening again."

https://www.vice.com/en/article/zmz3yw/why-is-my-google-doc-...

https://mashable.com/article/google-docs-locking-people-out

https://www.youtube.com/watch?v=JBgug0Fuwxgf

I work at Google. I’m not involved with this change specifically, but suffice it to say there is far more advanced processing than simple hash checks for CSAM being run on every file uploaded to Drive. I would be surprised if all files weren’t subject to these checks, but whether or not Google will take action on them is another matter.

I hope you are wrong as that would be economically and environmentally wasteful.

This is a point most miss. I’m curious how they know if something is/isn’t CP.

Anyone have any experience in this realm? Do they have to use CP to compare against? Wouldn’t that be illegal?

Look up photodna. Basically they compare hashes of uploaded images to a database of known child porn.