What did AI do in Iran- and Russia-linked influence campaigns?

OpenAI says available evidence suggests that employees at the center may not have known they were working for a Russian group. The company did not identify a Russian government body or intelligence agency as directing the operation.

OpenAI calls the Iran-origin activity Bogus Bylines. The company says its operators pitched articles under seven fake journalist identities and generated social-media comments. OpenAI identified nearly 100 articles published or syndicated under those bylines across more than a dozen publications, with publication dates from July 2025 through October 2026. These are OpenAI’s counts and have not been independently audited. The company said it could not identify who was behind the Iran-origin activity.

What did ChatGPT contribute?

OpenAI says people involved in the Russia-linked operation used ChatGPT mostly to prepare internal reports, though they also used it for some campaign material. For the Iran-linked operation, it says users asked the chatbot to edit drafts to fit publication requirements, prepare pitches and refine internal reports, as well as generate social-media comments.

That account describes people using AI as part of broader operations, not automated campaigns. OpenAI says some material was not made with its models and that the Iranian campaign’s social-media comments drew little engagement.

The company also compares the methods to earlier influence efforts built around fake media identities, including the Alice Donovan identity in 2016 and 2017 and the PeaceData outlet in 2020. Those examples predate the operations in OpenAI’s report. They show that fabricated journalist identities and front media organizations were not new tactics, even if AI tools could assist parts of the work.

What have independent fact-checkers confirmed?

Lupa Media found that a purported report alleging Ecuadorian recruitment for Ukraine was fabricated and had not been published by the broadcaster it imitated. Bolivia Verifica debunked an audio claim that water service in La Paz would be cut during a state of emergency.

The fact-checks confirm that those claims were false and circulated. They do not establish who created them, whether they came from the accounts OpenAI identified, or how many people saw or believed them. The connection between the checked claims and the operations rests on OpenAI’s investigation, not independent attribution by the fact-checkers.

Do OpenAI’s ratings show that the campaigns persuaded people?

OpenAI rated the Russia-linked campaign Category 5 and the Iran-linked campaign Category 4 on the six-category Breakout Scale, which classifies observable spread across platforms and communities. Category 4 covers mainstream-media amplification; Category 5 covers amplification by high-profile people. The ratings are OpenAI’s assessments, not independent measurements of audience exposure or changes in belief.

OpenAI’s account does describe distribution, including the articles it says appeared across publications. But independent measurement of how many people encountered the material, whether they believed it, or whether it produced offline effects is not established. Nor do the sources establish which people or organizations directed or funded either operation, or whether publishers carrying the Iran-linked articles knew the bylines were fake.

The evidence supports a narrower conclusion: human operators used AI in parts of influence workflows, while independent evidence of audience exposure and persuasive effects remains limited.