OpenAI has published a long-running series of reports detailing how it detects and shuts down accounts using ChatGPT for covert influence campaigns, scams, and cyber operations. Taken together, the company's disclosures describe well over a dozen influence operations alone, alongside fraud networks and state-linked hacking activity.
According to OpenAI's own reports, many of the influence campaigns trace to China. The company banned likely PRC-origin accounts behind operations it named "Tech and Tariffs" (comments and cartoons about US trade policy), "Data Center Bandwagon" (posts criticizing US data centers and AI infrastructure), "Nine-emdash Line" (content on the South China Sea, Hong Kong, and US politics), "Uncle Spam," "Sneer Review," "Sponsored Discontent," and the long-running "Spamouflage" network.
Russia-linked activity includes "Stop News," which OpenAI describes as recidivist — banned, then back again — targeting Ukraine, the West, Africa, and the UK; "Doppelganger"; "Bad Grammar"; and German-language operation "Helgoland Bite" focused on Ukraine and NATO. Iran-linked operations include STORM-2035, which OpenAI says generated election content about the US, UK, Ireland, and Venezuela, and the IUVM content network. OpenAI also reported an Israel-origin operation it dubbed "Zero Zeno," plus election-related networks in Ghana and Rwanda and political commenting in the Philippines.
The reports go beyond propaganda. OpenAI says it disrupted a Cambodia-based scam operation running investment, romance, gambling, and impersonation schemes, along with separate "pig butchering" and fake-review job scams. It banned Russian-speaking accounts building malware loaders, evasion layers, and command-and-control infrastructure; Korean-language accounts working on phishing and credential theft; and accounts matching publicly reported North Korea-linked IT-worker schemes that faked remote job applications. Other bans covered PRC-linked surveillance planning and profiling of critics.
Why it matters: these disclosures are among the clearest public evidence of how state actors and criminals are actually using general-purpose AI — and of how much the policing of that abuse currently depends on the companies selling the tools.