Disrupting a covert Iranian influence operation

Source: OpenAI Date: 2024-08-16 URL: https://openai.com/index/disrupting-a-covert-iranian-influence-operation

Summary

OpenAI threat intelligence disclosure from August 2024 describing how the company identified and terminated accounts used by an Iranian influence operation — specifically the “Storm-2035” network — that was using ChatGPT to generate political content, translate articles, and create social media material targeting US voters ahead of the 2024 election. OpenAI published indicators of compromise and the types of content generated, as part of a broader pattern of AI platform threat disclosures that became standard practice across major labs in 2024.

Implications

AI-assisted influence operations are operational, not theoretical. The Iran disclosure confirmed that sophisticated state actors were using frontier AI tools for disinformation at scale — not experimental use, but operational deployment with specific electoral targeting. The content generated was described as low-engagement (the operation wasn’t very effective), but the infrastructure was real.

Platform threat disclosure as industry norm. OpenAI, Meta, and Google all published influence operation disclosures in 2024, establishing a pattern of voluntary disclosure that both creates public accountability and positions the platforms as responsible actors rather than passive infrastructure. The timing (August 2024, three months before US elections) was deliberate.

Thread: AI and information operations. Sits alongside the June 2025 malicious uses disruption report and the May 2024 source understanding post as OpenAI’s ongoing thread on how its models are being misused for influence and disinformation, and what they’re doing about it.

Watch: Whether the disruption reports correlate with actual detection and takedown of influence operations, or primarily serve as public relations that documents the problem without containing it.