Tort Report: Abusive reporting activity

This case study was originally published in OpenAI’s October 2024⁠(opens in a new window) report.

Actor

We banned a small number of accounts that mainly operated in Vietnamese and generated comments that appeared to be designed for the purpose of filing reports against posts by Vietnamese independent media outlets on Facebook and YouTube. None of the Facebook or YouTube posts that the accounts were targeting had been taken down as of the date of this report. Given that the activity appears to have been aimed at abusive reporting—similar to that described by Meta in December 2021—we have dubbed this activity, “Tort Report”.

Behavior

This activity focused on generating short comments that could be used to file social media reports against videos posted by independent Vietnamese media outlets on YouTube and Facebook. This does not appear to have included any effort to get our models to watch, transcribe or otherwise ingest the detailed content of the video: it focused on the video title and any description. Given our limited visibility, we do not have evidence to determine whether any of the reports was filed with Facebook or YouTube. Where we identified specific posts on Facebook and YouTube referenced by these reports, the posts were still live online, indicating that any report that may have been made was not effective.

Completions

The content that this operation generated consisted of short comments in English and Vietnamese, suitable for being filed as a social media report. Since the input was limited to the post title and text, not the whole video, the comments were typically generic, stating that the video in question violated the rules, but without naming evidence as to why.