OpenAI Safety has detailed the disruption of four influence operations, codenamed "Sneer Review," "Stop News," "STORM-2035," and "A2Z." These operations, originating from China, Russia, Iran, and an unspecified source respectively, utilized OpenAI's models, including ChatGPT and DALL·E, to generate content for social media and other platforms. The accounts were banned after detection of their use of AI for covert influence activities.
Key Points
- Operation "Sneer Review," detailed in OpenAI's June 2025 report, involved accounts primarily issuing prompts in Chinese to generate social media posts and internal performance reviews, focusing on political and geopolitical topics relevant to China.
- This operation generated short social media comments in English, Chinese, and Urdu, posted on platforms such as TikTok, X, Reddit, and Facebook, and targeted a Taiwanese social media influencer and Pakistani activist Mahrang Baloch.
- Operation "Stop News," described in OpenAI's October 2024 report, originated in Russia and used models to generate content in English, French, and Russian, including long-form articles and images created with DALL·E.
- The "Stop News" operation posted content on websites posing as news outlets in Africa and the UK, and on X, with some social media accounts featuring profile pictures generated by older generative adversarial networks (GAN).
- Operation "STORM-2035," also from OpenAI's October 2024 report, originated in Iran and generated long-form web articles and short comments in English and Spanish, focusing on US and UK election content, the conflict in Gaza, and other political topics.
- The "STORM-2035" operation posted content on X and Instagram, with some accounts posing as partisans of US presidential candidates or supporters of Scottish independence.
- Operation "A2Z," detailed in OpenAI's October 2024 report, used the API to create comments for X and Facebook, often praising Azerbaijan or defending its human rights record, and leveraged AI to manage fake social media personas and draft multilingual replies.
Context
According to OpenAI Safety, these operations represent varied approaches to using AI for influence. "Sneer Review" focused on creating a false impression of organic engagement through reply chains and targeted specific individuals and topics aligned with China's geostrategic interests. "Stop News" was noted for its prolific use of DALL·E-generated imagery and its establishment of "information partnerships" with public entities in the UK.
"STORM-2035" engaged in parallel workstreams generating both long-form articles for fake news websites and short comments for social media, sometimes using AI for basic research on social media engagement. "A2Z" demonstrated the use of AI to manage numerous social media accounts and engage in conversations with real users, highlighting the efficiency AI can bring to such operations.
Why It Matters
The detailed reports from OpenAI Safety provide insights into how AI models are being leveraged for influence operations, demonstrating the evolving landscape of inauthentic activity. Builders and practitioners must note the specific techniques and platforms used by these operations to better understand and mitigate potential misuse of AI technologies.
What To Do
- Watch for OpenAI's future threat reports, such as the June 2025 and October 2024 publications, for updated information on influence operations.
- Note the specific platforms and content types identified in these reports (e.g., TikTok, X, Reddit, Facebook, DALL·E images) as potential vectors for AI-generated influence.
- Review the descriptions of how AI was used to generate content, manage personas, and simulate engagement to inform defensive strategies.
- Consider the varying levels of sophistication and engagement observed across these operations when assessing potential threats.
Keep Exploring
/atlas/claude-family /atlas/gpt-family /atlas/image-models