OpenAI banned a cluster of accounts for the second known round of a covert influence operation.
The company says the banned accounts were tied to STORM-2035, an operation it assesses is likely linked to Iran. The accounts used OpenAI's models to generate political content aimed at audiences in the US, UK, Ireland, and Venezuela. OpenAI's own label for this takedown - "recidivist influence activity" - signals repeat offenders rather than a first-time bad actor. The report doesn't detail volume, reach, or specific narratives pushed this round.
The recurrence is the real story here. A ban is a one-time cost for an operator who can just spin up new accounts; detecting and disrupting them is a permanent, unpaid tax on the AI labs. That asymmetry is exactly what has kept human-run sockpuppet farms alive for a decade, and it's now showing up in generative AI accounts too.
Doppelganger and Spamouflage-style networks proved this playbook works on social platforms for years - STORM-2035 just confirms it survives the jump to AI chatbots just fine.