OpenAI traced seven fake journalists and roughly 100 placed articles to an Iran-based operation
The vendor’s disclosure, made last week and written up on Monday, found a 15-month run of ghost bylines into small and syndicated outlets — and assessed the operators were commercial actors for hire, not the Iranian state.
According to the Jerusalem Post’s 12 October account of the OpenAI report, an Iran-based covert influence operation used AI tools to generate social media comments, built at least seven fictitious journalist personas and pitched long-form articles to outlets internationally. OpenAI identified nearly 100 articles produced by the operation that were published in syndicated publications and small and medium-sized outlets worldwide, the earliest dating to July 2025 and the latest to October 2026. Volume rose after the outbreak of the US–Iran conflict, with content focused primarily on that war and a smaller share on US domestic politics. Several personas maintained accounts on X, Facebook, Instagram or Threads, which OpenAI assessed were operated from inside Iran or elsewhere in West Asia. The report found none of the AI-generated social media comments attracted substantial engagement.
Two caveats belong in any account of this. The disclosure is not Monday news: JPost says OpenAI announced it “on Thursday,” which places the report at 8 October, and an NCRI summary of the same investigation is dated 10 October. NCRI is the political wing of the MEK and is useful here only for establishing that the report exists. The second caveat is sampling. OpenAI’s visibility comes from its own platform telemetry, meaning the operations it finds are the ones that used its tools — a self-selected sample that says nothing about campaigns run elsewhere. OpenAI’s own attribution assessment, based on overall activity patterns, was that the operators were likely commercial actors running a for-hire campaign rather than an arm of the Iranian state.
Assessment: The finding that matters is the inversion: the operation succeeded at placement and failed at amplification. Comments went nowhere; bylines got printed. That points the defensive problem away from platform moderation and toward editorial intake at small outlets and syndication wires, which have no capacity to vet a freelance pitch from a persona with a plausible X account. The for-hire assessment will be flattened into “Iranian state propaganda” by most coverage this week; it is not what OpenAI said. The open question is whether OpenAI named the publications that ran the roughly 100 articles. If it did, there is a second story about which mastheads carried ghost copy — and whether any of them were in the Gulf.