OpenAI has identified and banned accounts linked to a network it dubbed "Fish Food," which utilized AI to support multilingual influence operations. This activity, detailed in OpenAI's February 2026 report on disrupting malicious AI uses, involved generating content for websites and social platforms.
The accounts were connected to the "Rybar" network, active on Telegram and X, with some accounts likely originating in Russia. The operation used AI, including ChatGPT and Sora, to produce content in languages such as Russian, English, and Spanish, which was then disseminated online.
Key Points
- OpenAI banned ChatGPT accounts linked to the "Rybar" network, some likely originating in Russia.
- The operation, dubbed "Fish Food," used AI to generate multilingual content for influence activities.
- Content was posted on "Rybar"-branded accounts and other social media accounts without declared connections.
- One user employed ChatGPT to develop commercial plans for covert interference campaigns in Africa.
- The network also used ChatGPT to translate services "Rybar" could offer, including running social media accounts and an "investigative journalism" website.
- A user asked ChatGPT to edit a proposal for an election interference team, apparently in Africa.
- Sora was used by one user to generate videos promoting the "Rybar" brand.
- The content generated typically praised Russia and its allies while criticizing Ukraine and Western countries.
Context
According to OpenAI, the primary activity detected was content generation for social media. Users typically prompted in Russian but generated output in multiple languages. This content was then posted on "Rybar"-branded social media accounts and its main website, as well as by other X and Telegram accounts that did not declare a connection to Rybar. OpenAI noted that the ChatGPT activity functioned as a content farm for these accounts.
Why It Matters
This case demonstrates how AI models can be leveraged to scale influence operations, posing challenges for platform providers in identifying and mitigating such misuse. It highlights the need for ongoing vigilance and policy enforcement by AI developers to prevent their tools from being used for deceptive or manipulative purposes.
What To Do
- Note the February 2026 publication date of OpenAI's report on malicious AI uses.
- Watch for further case studies from OpenAI regarding disruptions of malicious AI activities.
- Review OpenAI's safety policies concerning the use of their models for influence operations.
- Consider how AI-generated content can be disseminated through both branded and unbranded channels.
Keep Exploring
/atlas/claude-family /atlas/gpt-family /atlas/**grok**-family /atlas/**gemini**-family /atlas/llama-open /atlas/image-models