OpenAI Exposes Russian Op Using ChatGPT to Boost Germany's Far-Right AfD

OpenAI has pulled back the curtain on a clumsy but telling Russian influence operation that used its own models to churn out German-language political propaganda. The company banned a cluster of accounts originating from Russia for what it dubbed Operation “Helgoland Bite” — a campaign that generated content supporting the far-right Alternative für Deutschland (AfD) party while criticizing Ukraine, NATO, and the US ahead of Germany’s 2025 election.

The operation’s hub was a Telegram channel called “Nachhall von Helgoland,” a phrase nodding to a North Sea island. It posed as an independent local German news outlet but scraped together only 1,755 subscribers. The real velocity came from cross-posting. OpenAI confirmed the channel’s AI-generated articles were republished verbatim on a German-language site within the Pravda network, a known node in the Kremlin-linked “Portal Kombat” disinformation ecosystem previously flagged by French intelligence. An X account with over 27,000 followers further amplified the content, complete with an AI-generated profile picture.

The behavior went beyond drafting articles. The banned accounts asked ChatGPT for publicly available contact details on German opposition activists and bloggers. They also fed the model Russian text for translation into German, revealing snippets of internal coordination — messages about scheduling social media posts and even references to payments. It’s the kind of operational chatter that shows these weren’t just random trolls experimenting, but a structured effort to use AI as a production pipeline.

On the Breakout Scale for information operations, OpenAI pegs this at the upper end of Category 2: multi-platform activity that ultimately fizzled, generating minimal authentic engagement. That’s a familiar pattern. These operations rarely go viral on their own; they need a real audience to catch the bait. The 27,000 followers on X looks impressive on paper, but follower counts are a vanity metric if you’re shouting into a void of bots and ideologues. What’s notable here is the audacity of using the very tools Silicon Valley builds to attack Western political cohesion — and the fact that the generated text was apparently good enough for the Pravda network to run without a human rewrite.