OpenAI bans Vietnam-based accounts that gamed reporting systems with AI slop

OpenAI has banned a cluster of accounts from Vietnam for a lazy but telling form of platform manipulation they’re calling “Tort Report.” The operation was dead simple: use OpenAI’s models to generate short, generic comments that could be filed as abuse reports against independent Vietnamese media outlets on Facebook and YouTube. No sophisticated deepfakes, no complex social engineering. Just AI-generated text designed to gum up content moderation machinery.

The accounts produced comments in both English and Vietnamese that vaguely claimed videos violated platform rules, but they conspicuously lacked any specific evidence. That’s because the system never actually watched the videos. It only ingested titles and descriptions—a telling shortcut that made the output easy to spot. “The comments were typically generic,” OpenAI noted in its October 2024 influence operations report, where this case study first appeared. The whole scheme reads less like a cyber-espionage thriller and more like someone feeding a few text prompts into a bot and hitting submit.

Here’s the silver lining, if you can call it that: the operation was a complete flop. Every single Facebook and YouTube post targeted by these AI-generated reports remained live as of the report’s publication date. OpenAI assessed the impact at Category 1 on its Breakout Scale—the lowest rung, indicating zero observable effect on either platform. That’s almost worse than failing. It’s failing while leaving a trail of obviously synthetic text that makes the perpetrators look incompetent.

Still, the incident highlights a grinding, unsophisticated threat that platforms will increasingly face. This wasn’t a state-of-the-art influence campaign. It was the equivalent of stuffing a ballot box with photocopies. Meta described similar abusive reporting tactics back in December 2021, and the fact that actors are now reaching for off-the-shelf AI to scale the drudgery suggests a future where moderation teams spend less time fighting master propagandists and more time swatting away automated paper-pushers. The real risk isn’t that these campaigns work. It’s that the sheer volume of synthetic noise makes it harder to spot the attacks that actually matter.