Persistent! Russia's 'Stop News' Network Returns with AI-Generated Videos

Hi, it's Shiichan! Today's story is a little surprising and not exactly funny, so let me walk you through an OpenAI report on countering misuse of AI.

OpenAI News openai.com

What was announced?

OpenAI's News page reported that the Russia-origin influence network it named and disrupted as "Stop News" back in October 2024 had tried to get back into action. OpenAI banned a set of accounts belonging to the same network that were trying to generate new content with ChatGPT. The operation still appears consistent with being run by a marketing company, just like the original investigation found.

The story so far

Stop News was the operation OpenAI originally disrupted in October 2024. It mass-generated news articles and social media comments in English, French, and Russian, then posted them across websites posing as news outlets in Africa and the UK, plus platforms like X. It was also unusually heavy on DALL·E-generated cartoon images, and had landed "information partnerships" with a school in Wales and a church in Yorkshire. Based largely on that apparent real-world reach, OpenAI originally rated the operation Category 3 on the Breakout Scale: activity across multiple platforms with some involvement from real organizations and individuals.

What changes

This time, OpenAI found the same network trying to keep generating content with ChatGPT and banned the accounts involved. The key difference from before is that the operation shifted its focus toward video content rather than just text and images. The operators had ChatGPT write Russian-language scripts, translate them into French, and then generate SEO-optimized descriptions and hashtags, one continuous pipeline. The videos themselves, featuring an AI newsreader, weren't made with OpenAI's models; the operation appears to have used another company's AI tools, though OpenAI couldn't independently confirm which service.

Dive Deep

The new Stop News playbook turned out to be fairly elaborate.

• Operators had ChatGPT write long Russian-language scripts, split them into two-sentence snippets, and generate a video prompt for another AI model for each snippet
• They then had the full script translated into French. One video's caption contained a French grammar error mixing up "ces" and "ses", even though ChatGPT's own translation was correct, suggesting the caption was transcribed from audio without a French speaker proofreading it
• Scripts turned up on a YouTube channel tied to newstop[.]africa, the operation's Africa-focused website, as well as on YouTube and TikTok channels that didn't carry the "Stop News" brand at all. The AI newsreader's appearance on these channels visibly changed between January and June 2025
• Content fell into three main buckets: French-language material criticizing France's and the US's role in Africa while praising Russia, English-language material criticizing Ukraine and its supporters, and promotional copy for Russian commercial services, including ads for unlicensed online gambling
• Image generation dropped sharply compared to the original operation, matching an observation from the French agency VIGINUM that AI-generated image use fell drastically industry-wide after the OpenAI and Meta reports were published

Reach stayed limited overall. The core "Newstop Africa" X account had just 172 followers as of August 2025, and its most-retweeted post only got four retweets. The AI-newsreader YouTube and TikTok channels each had around 1,900 followers; the TikTok channel racked up 5,855 total likes across 56 videos, with its most-viewed video hitting 63,300 views and its least-viewed just 87. The YouTube channel logged about 255,000 total views across 50 videos, averaging 5,100 per video, with no sign the videos were re-shared or cited elsewhere.

What's especially notable is that OpenAI revised its earlier assessment. In 2024, the Category 3 rating rested largely on those apparent "information partnerships" with UK websites. Subsequent research by VIGINUM and open-source researchers showed those partnerships were likely fictional: the operation had exploited technical flaws on the partner sites to insert content without the site owners' knowledge. Based on that, OpenAI now assesses this operation as Category 2, activity across multiple platforms without meaningful breakout to authentic audiences.

Wrap-up

• OpenAI banned new ChatGPT use by the same Russia-origin network behind the 2024 "Stop News" operation
• This time the operation leaned into video scripts, translation, and SEO text generation rather than images; the videos themselves used another company's unidentified AI tools
• It posted through AI-newsreader YouTube/TikTok channels and newstop[.]africa-linked X accounts, but followers, views, and spread all stayed low
• The 2024 "information partnerships" with UK sites turned out to be fabricated, exploiting technical flaws rather than real cooperation, so OpenAI downgraded its Breakout Scale rating from Category 3 to Category 2
• Worth a read if you're tracking AI safety, this is a real example of a disrupted operation trying to come back in a different form