Artificial intelligence is opening a new front in online influence operations, with Iran, China and private groups in Israel using autonomous AI agents to run social media campaigns at a scale that would have been difficult to achieve with human operators alone.
U.S. officials and security researchers familiar with the operations said the campaigns used increasingly capable Chinese open-source AI models to create autonomous agents capable of performing tasks across social media platforms with limited human involvement.
The agents were deployed in large numbers to create networks of fake accounts on platforms including Instagram, Facebook, X and TikTok. They then populated those accounts with politically charged and misleading content aimed at shaping public opinion and intensifying existing divisions.
In several cases, safeguards that would normally prevent AI systems from creating fake accounts or manipulating online discussions were disabled by the operators, according to U.S. officials who were not authorised to speak publicly about sensitive intelligence.
The activity has drawn the attention of U.S. intelligence agencies, technology companies and cybersecurity researchers, who identified the Chinese and Iranian operations as state-backed. Two other campaigns were linked to private Israeli companies.
The Iranian operation raised particular concern because it targeted U.S. audiences through accounts designed to appear as ordinary Americans living in major cities.
The AI-generated accounts interacted with journalists and politicians and posted material that included popular internet memes and content critical of the Republican Party. Nearly 80,000 people followed the accounts during the first half of the year, according to officials and researchers familiar with the campaign.
The operations demonstrate how the combination of open-source AI models and autonomous agents could change the economics of influence campaigns. Instead of relying on teams of operators to create accounts, write posts and engage with users individually, automated systems can perform many of those tasks simultaneously.
One of the Israeli-linked campaigns was created during the summer by IntelEye, a Tel Aviv-based company, according to people familiar with the operation.
Maor Sellek, an IntelEye co-founder, said the activity was an experiment designed to test the safety of AI models. He denied that the company operates influence campaigns.
The emergence of these operations is adding to concerns among security researchers and government officials that AI-powered influence activities could become faster, cheaper and harder to detect as autonomous systems become more capable.
The use of open-source models is also complicating efforts to impose safeguards. Unlike tightly controlled commercial AI systems, open-source models can be downloaded, modified and deployed by users with fewer restrictions.
For governments and technology companies, the challenge is increasingly shifting from detecting individual fake posts to identifying automated networks capable of creating and managing entire influence operations with minimal human intervention.
