OpenAI says it stopped multiple covert influence operations that utilized AI models
OpenAI recently revealed that it successfully halted five covert influence operations that were leveraging its AI models for deceptive activities across the internet. These operations, which were dismantled between 2023 and 2024, were traced back to countries like Russia, China, Iran, and Israel. The primary aim of these operations was to sway public opinion and influence political outcomes clandestinely, without disclosing their true identities or motives.
The company stated in a statement released on Thursday that, "As of May 2024, these campaigns do not appear to have meaningfully increased their audience engagement or reach as a result of our services." OpenAI collaborated with various stakeholders from the tech industry, civil society, and governments to put an end to these malicious activities.
Impact of Generative AI on Elections
OpenAI's report comes at a crucial time when concerns are mounting about the influence of generative AI on upcoming elections worldwide, including those in the US. The investigation conducted by OpenAI shed light on how networks of individuals involved in influence operations exploited generative AI to create vast amounts of text and images, along with generating fake engagement by deploying AI to fabricate comments on social media posts.
Ben Nimmo, the principal investigator at OpenAI's Intelligence and Investigations team, highlighted the significance of the report during a press briefing, stating, "With this report, we really want to start filling in some of the blanks."
Details of Covert Operations
One of the identified covert operations by a Russian group, dubbed as "Doppelganger," employed OpenAI's models to craft headlines, transform news articles into Facebook posts, and produce comments in multiple languages to undermine support for Ukraine. Another Russian entity utilized OpenAI's models to debug code for a Telegram bot that disseminated brief political comments in English and Russian, targeting regions like Ukraine, Moldova, the US, and Baltic States.
Similarly, the Chinese network known as "Spamouflage," recognized for its influence efforts on platforms like Facebook and Instagram, leveraged OpenAI's models to analyze social media trends and generate text-based content in different languages on various platforms. Additionally, the Iranian group named "International Union of Virtual Media" also utilized AI to generate diverse content in multiple languages.
OpenAI's disclosure aligns with the transparency efforts of tech companies to address such deceptive practices. For instance, Meta recently published its latest report outlining coordinated inauthentic behavior involving an Israeli marketing firm that orchestrated an influence campaign targeting individuals in the US and Canada on Facebook.
