OpenAI Exposes Russian and Iranian Influence Operations Using AI

OpenAI has revealed that Russian operatives exploited its AI chatbot, ChatGPT, to update superiors on influence campaigns across several countries. These efforts were previously unattributed to Russia. OpenAI’s investigation led to banning Russian and Iranian accounts involved in using its AI tools for such operations.

According to OpenAI, one significant campaign managed to disseminate false narratives through established news outlets. This campaign even prompted a response from a European Parliament member based on internal reports created using ChatGPT. The company’s analysis showed that campaigns focusing on embedding false narratives in media outlets reached a broader audience compared to those using fake social media accounts.

The Russian operations targeted Latin American countries with the goal of reducing support for Ukraine and impacting local politics. Accounts mostly utilized ChatGPT for writing reports to unspecified higher-ups. One report described an extensive influence operation that prompted public response from a political figure. Russian operatives impersonated a regional educational authority in Peru, tricking schools into organizing activities about Ukraine featuring Stepan Bandera, a divisive historical figure.

These activities stirred tensions between Ukrainians and Poles, leading a far-right Polish MEP to suggest banning individuals supporting Bandera. “Russians are constantly engaging in these sorts of active measures,” noted Darren Linvill from Clemson University, highlighting the use of narrative laundering tactics.

The Russian memos also outlined the creation of a fake think tank in Latin America, recruiting unwitting researchers to conduct interviews and write reports. OpenAI linked this to previous Russian operations pretending to be an independent news outlet associated with Yevgeny Prigozhin’s Internet Research Agency.

In some cases, Russian operatives claimed credit for events they did not orchestrate. AI played a minor role in generating fake material within these efforts. Nonetheless, using AI as a shortcut led to exposure of their activities, according to Linvill.

In parallel, Iranian operatives used AI to distribute content in Persian and English, mainly focusing on the U.S.-Iran conflict. OpenAI noted that Iranian accounts created seven fake authors to write and polish articles for small global media outlets. Nearly 100 articles were published under false personas in outlets like the Daily Kos and Middle East Monitor. These sites have since removed the fake authors’ pages, though some content remains. The Daily Kos noted that its community section allows anyone to post, with few pre-publication reviews.

Despite this, Linvill observed that the AI’s ability to generate authentic-looking language was invaluable to these operations, enabling them to publish in legitimate outlets. Without AI, such operations would have faced greater challenges due to language barriers.