Operation “A2Z”: Multilingual influence activity
OpenAI ● Covered by 2 sources
OpenAI shut down a covert network called "A2Z" that used its AI to churn out political content in multiple languages. It targeted elections, Ukraine, and hot-button politics across social platforms.
Based on reporting by OpenAI — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
OpenAI's latest threat report names a new offender: a covert influence operation the company has dubbed A2Z. Investigators found a cluster of accounts leaning on ChatGPT to mass-produce political content in several languages at once, then spreading it across various social platforms.
The targets weren't random. A2Z zeroed in on elections, the war in Ukraine, and other divisive political topics — the kind of subjects where a flood of seemingly organic posts can nudge public sentiment without anyone tracing it back to a single source. Using AI to translate and localize propaganda isn't new, but the scale and speed these tools now offer changes the math for anyone running an influence campaign. What used to require a team of writers in multiple languages now takes one operator and a chatbot.
OpenAI says it identified the network's behavior patterns and cut off the associated accounts, folding the case into its ongoing effort to track and disrupt misuse of its models for coordinated deception. The company has been publishing these takedown reports periodically, and each one reads like a snapshot of how state-linked or opportunistic actors are experimenting with generative AI for propaganda, phishing, and now, apparently, election-adjacent narrative-shaping.
What's notable here isn't that someone tried this — it's that OpenAI caught it and named it. Multilingual influence ops used to be a specialty skill set. Now they're a prompt away, and the company policing its own platform is, for the moment, the main line of defense against that becoming routine.
My take — AI-written commentary, not fact-checked reporting
This is the pattern to watch: AI didn't invent influence operations, it just made them cheaper and faster to run in five languages at once. I'll take OpenAI's transparency here as a genuine plus, but let's not pretend platform-level whack-a-mole is a real solution — it's a stopgap while regulators and social networks catch up to a threat that's scaling faster than the takedowns.
Read more about this at: OpenAI
