Operation “Helgoland Bite”: German-language influence activity
OpenAI banned accounts that appeared to originate in Russia and used AI to generate German-language political content about Ukraine, NATO, and domestic issues.
This case study was originally published in OpenAI’s June 2025(opens in a new window) report.
Actor
We banned ChatGPT accounts that appeared to originate from Russia. They were using our models to generate German-language content about the German 2025 election, and criticizing the US and NATO. The content was distributed on Telegram and X.
Behavior
Through off-platform investigations, we were able to confirm associated generations were deceptively distributed in a Telegram channel named “Nachhall von Helgoland”(opens in a new window) (Echo of Helgoland, referencing an island in the Heligoland Bight of the North Sea). This channel, described as locally operated independent German news, had 1,755 subscribers at the time. Given the name, and the campaign’s attempt to generate critical commentary, we have dubbed this operation “Helgoland Bite.”
The content from this channel was regularly reposted verbatim on a domain affiliated with the Pravda network(opens in a new window) targeting German-speaking audiences. The associated Pravda (DE) website is a known node in the Moscow-linked covert influence operations network “Portal Kombat,”(opens in a new window) previously identified by the French government’s VIGINUM service.
Article on the Pravda DE website, sourced from Nachhall von Helgoland. The headline reads, “Orbán meets Weidel: ‘AfD is Germany’s future.’”
The network also distributed content generated by our model via an X account with over 27,000 followers, frequently posting AI-generated content that supported the Alternative für Deutschland (AfD) party, and using an AI-generated profile image.
Tweet generated by this threat actor using our models. The profile picture was also generated using our models. The tweet reads, “We urgently need a ‘DOGE ministry’ when the AfD finally takes office. The first thing it should do is evaluate politicians for their suitability and, if necessary, reclaim expenses if they prove unfit. I think that could bring in a lot of money.”
Completions
As well as generating short articles and social media comments, the accounts asked our models for publicly available information about German opposition activists and bloggers, including ways to contact them. They also asked for short texts to be translated from Russian into German, where the language usage was consistent with messaging or conversational traffic. Some of these messages appear to have discussed coordination of posting times for social media content; others referenced payments.
Impact
As noted above, the Telegram channel counted 1,755 subscribers at the time of our investigation, and was regularly reposted verbatim on a domain affiliated with the Pravda network. The X account had over 27,000 followers.
Using the IO impact Breakout Scale(opens in a new window), we would assess this as being towards the upper end of Category 2: activity on multiple platforms, but little authentic engagement or evidence that their content was widely shared.