Hundreds of AI Agents Drove Multi-Platform Influence Campaigns

This is less a story about fake content than about agent autonomy. The newly documented campaigns show AI agents executing multi-step workflows — creating personas, drafting posts, tagging real journalists and politicians, and coordinating across four platforms — with far less direct human involvement than any prior operation. For AI researchers and builders, that shift from generative output to autonomous orchestration raises hard questions about evaluation, safety, and who is accountable when an agent, not a human operator, drives the campaign.

A New York Times investigation, syndicated by the Fact Check Team across local outlets on September 21, 2026, documents a threshold shift in online influence operations: Iran, China, and private Israeli firms deployed AI agents — software systems that execute multi-step tasks with minimal direct human involvement — to run social media influence campaigns. The agents handled the full operational stack, including creating fictitious accounts, generating political and current-events content, and coordinating activity across Facebook, Instagram, X, and TikTok. This is the core development: influence operations are moving from labor-intensive human work to largely automated, orchestrated software pipelines.

“

officials have previously warned that foreign actors, including Iran, have used generative AI and inauthentic personas to attempt to influence Americans and sow discord.

AI-generated content and fake online personas are not new. U.S. officials have previously warned that foreign actors, including Iran, have used generative AI and inauthentic personas to attempt to influence Americans and sow discord. What researchers say is genuinely different about these newer campaigns is the degree of automation. Earlier influence efforts — most notably the Russian Internet Research Agency's troll farms — required buildings full of human operators who manually created accounts, wrote posts, and coordinated messaging. In the newly identified campaigns, some operations used hundreds of AI agents, dramatically reducing the human labor and cost previously required to run a credible influence network at scale.

The Iranian operation illustrates the new playbook. Its AI-generated accounts presented themselves as ordinary Americans living in major U.S. cities, posted memes and political commentary, tagged journalists and politicians, and promoted messaging critical of the Republican Party. These accounts accumulated nearly 80,000 followers during the first half of 2026. That figure, however, should not be read as 80,000 Americans persuaded. The reporting establishes reach — the number of followers — but does not establish how many people believed the content or changed their political views as a result. This distinction between reach and persuasion is a crucial attribution gap for researchers, platforms, and policymakers; without better measurement, there is a real risk of both overstating and understating the operational impact of agentic influence campaigns.

For platforms and security teams, the implications are substantial. Automation erodes the cost barrier that previously constrained influence operations, acting as a force multiplier that lets a small number of operators field large networks of seemingly authentic personas. Cross-platform coordination compounds the detection problem: any single platform sees only a fraction of a campaign's total activity, because the agents distribute account creation, posting, and amplification across Facebook, Instagram, X, and TikTok simultaneously. Trust-and-safety systems and bot-detection models tuned for manual or semi-automated behavior must now contend with agents that mimic authentic posting patterns, maintain coherent personas, and coordinate across ecosystems — a detection challenge that spans technical, policy, and legal boundaries.