OpenAI Highlights Russian and Iranian Influence Operations Using AI Chatbot
OpenAI announced that it banned Russian and Iranian accounts engaged in using its AI tools for influence campaigns in various countries. These operatives reportedly placed false narratives in media outlets, even prompting responses from political figures, according to internal reports created using OpenAI’s ChatGPT.
Investigations by OpenAI revealed that efforts to seed false narratives in established media had greater impact than attempts to manipulate through fake social media accounts. Russian operations focused on Latin America, aiming to undermine support for Ukraine and manipulate local politics. These accounts primarily used ChatGPT to write reports for their superiors.
One significant campaign led to a public response from a European Parliament member. The operatives pretended to be a regional educational authority in Peru, deceiving schools into organizing activities about Stepan Bandera, a controversial historical figure.
Stepan Bandera led the Ukrainian independence movement before and during WWII. While nationalists admire him, others, including Poles, condemn his alliance with Nazis against Poles and Jews.
This led a Polish MEP to propose banning Bandera supporters. The campaign exploited historical tensions between Ukrainians and Poles.
Darren Linvill of Clemson University’s Media Forensics Hub highlighted these operations as typical Russian attempts at narrative manipulation through propaganda and false personas. Russian internal memos mentioned a fake Latin American think tank and unwitting researchers employed for reports. This resembled a 2020 Russian effort posing as an independent news outlet linked to Yevgeny Prigozhin’s Internet Research Agency, active during the 2016 U.S. election.
While AI-generated fake material was minimal, its use helped catch these activities. Linvill noted Russians tried using AI to ease their tasks, similar to general AI use, but this contributed to their exposure.
Iranian operatives used AI to disseminate content in Persian and English, focusing on U.S.-Iran tensions and, to a lesser extent, American politics. This campaign resembled commercial actors running for-hire influence operations, though OpenAI couldn’t identify the actors. Iranian accounts created seven fake authors to present long articles globally.
Fake authors published in various outlets, including the progressive U.S. news site Daily Kos and the U.K.-based Middle East Monitor. NBC News reported that the Middle East Monitor removed fake author pages after identifying 21 articles by four such personas. Daily Kos clarified the fake persona posted in its community section, independent of staff-reviewed reports.
Linvill emphasized AI’s role beyond shortcuts, facilitating authentic-seeming language for publication in legitimate platforms. A few years ago, Iranian operations might have struggled more with this task.