Iran and China have used open-source AI systems to automate fake accounts, content and online engagement, raising fresh concerns about the technology’s role in political influence.
Iran, China and Israel-linked groups have used AI agents to create fake social media accounts and automate influence campaigns, a New York Times report said.
- AI agents run influence operations with limited human input.
- Iran, China, Israel groups used AI for social media campaigns.
- AI agents create fake accounts, post, and coordinate activity.
Artificial intelligence is moving beyond creating fake images and posts and into running influence operations with limited human involvement. Iran and China, along with groups in Israel, have used AI agents to build and operate networks of social media accounts in recent months.
According to the New York Times, the campaigns combined openly available Chinese AI models with “agents”, software systems designed to carry out tasks on their own. Hundreds of these agents were deployed to create fake accounts, publish posts and coordinate activity across platforms including Instagram, Facebook, X and TikTok.
The approach allowed operators to run campaigns at a scale that would have been difficult to achieve manually. The agents could handle multiple stages of the process, reducing the need for people to manage every account or post.
Fake accounts at scale
US officials and security researchers said the Iranian and Chinese efforts were state-backed, while two Israeli campaigns were linked to private companies. The campaigns drew attention because AI agents were being used not simply to generate material but to manage networks and distribute it.
Iran’s campaign focused on US audiences. AI-created accounts posed as ordinary Americans living in major cities and posted political material, including memes and content critical of Republicans. The accounts also tagged journalists and politicians in comments.
Nearly 80,000 people followed the AI-created accounts, which were active during the first half of the year.
Details about the Chinese campaign were more limited. Officials said it experimented with the same basic approach of using AI agents and open-source models to automate influence activity.
Open models raise new risks
Open-source AI models make the technology easier to adapt because they can be downloaded, modified and run on private hardware. That also means the companies that developed the models may have less ability to intervene when their systems are misused.
Security researcher Kyle Chan said open models could create a new level of autonomy compared with closed systems operated by companies such as OpenAI and Anthropic.
The latest campaigns have added to wider concerns about AI safety. Researchers and US officials are watching how quickly automated agents are becoming capable of acting across multiple digital platforms without continuous human control.
AI has already been used in election-related influence operations, including campaigns involving Russia, Iran and China. Officials now fear newer systems could make such efforts easier to launch and harder to detect.
Israel-linked experiments
One Israeli campaign was created by IntelEye, a Tel Aviv-based company founded by former NSO Group employees. The operation used DeepSeek to control thousands of social media accounts and interact with posts about Israel’s upcoming national elections.
IntelEye co-founder Maor Sellek said the activity was a safety experiment intended to understand how poorly protected AI models could be abused. He acknowledged that the company should have informed social media platforms about the testing.
Meta said it had detected and removed most of the accounts linked to the operation, describing it as a coordinated effort to create inauthentic accounts across several platforms.
The experiments show how influence campaigns are changing as AI agents become more capable. What once required teams of people can increasingly be divided into automated tasks, creating networks that can produce, adapt and distribute political content with little direct human involvement.
