State-backed actors and private firms deploy hundreds of bots to create fake accounts and spread disinformation ahead of US midterms

The flags of China and the United States. [AFP]
The flags of China and the United States. [AFP]

State-backed actors from Iran and China, along with private Israeli companies, have deployed AI agents to conduct what amounts to autonomous influence operations on social media, The New York Times reported Friday (local time).

The AI systems created fake accounts, generated politically charged content and spread disinformation — carrying out the full cycle of influence operations that humans once managed, with minimal human involvement. The reports have heightened concerns that AI-driven online interference could intensify ahead of the US midterm elections in November.

Citing US officials and cybersecurity experts, The New York Times said the actors involved rapidly expanded their online influence operations using AI at speeds and scales that surpass human capacity, with only minimal human oversight.

They built AI agents — bots capable of autonomously executing specific tasks — using open-source Chinese AI models that can be freely downloaded and modified.

Hundreds of AI agents autonomously created fake accounts across platforms including Instagram, Facebook, X (formerly Twitter) and TikTok, then rapidly spread false posts on political and current-affairs topics through those accounts to manipulate public opinion.

The Chinese and Iranian campaigns were state-directed, while the Israeli campaign was traced to a private company. The Iranian campaign targeted the American public, with agent-generated fake accounts posing as ordinary Americans living in major cities.

Those accounts tagged journalists and politicians and spread popular memes and anti-Republican content. The accounts amassed roughly 80,000 followers through activity in the first half of this year alone.

Meta, which operates Facebook and Instagram, had separately announced late last month that it had disrupted an Iranian influence operation that used AI.

Officials said the cases represent what is effectively the first known instance of AI agents handling every stage of an influence operation — from account creation to the coordination of online messaging.

Although the AI agent-based activity was described as crude, it drew the attention of US intelligence agencies, technology companies and security researchers, raising alarm that publicly available AI models could be exploited to conduct online influence operations largely on their own.

Some US officials warned that similar attempts could be made targeting American voters ahead of the November midterms.

Kyle Crighton, a researcher at Georgetown University's Center for Security and Emerging Technology, predicted that AI agent-driven influence operations would become more common as the technology continues to advance. "Agents can become quite sophisticated, gaining the ability to blend in naturally like real human users — going well beyond simple scaling," he told The New York Times.


th5@heraldcorp.com