An Israeli media report has revealed growing US concerns over an Iranian “robot army” being deployed in cyber warfare, as artificial intelligence allows influence campaigns to operate on social media with almost no human involvement.
According to Yedioth Ahronoth, US intelligence and security officials are increasingly concerned about what they regard as a dangerous qualitative leap in cyber warfare. For the first time, countries including Iran and China, alongside private Israeli companies, have begun operating fully autonomous artificial intelligence campaigns across social media platforms.
These new “robot armies” no longer require teams of human operators. The systems can independently create fake accounts, generate content, interact with users and provoke public debates across X, Facebook, Instagram and TikTok with almost no human intervention.
The development, first reported by The New York Times, has raised concern in Washington, where officials fear the technology could eventually be deployed on a large scale to influence American public opinion ahead of the presidential and congressional midterm elections in November.
Foreign influence operations previously required entire teams of human “trolls” or employees to write content and manage accounts. The combination of open-source artificial intelligence models and autonomous AI agents now allows influence operations to be conducted on an unprecedented scale with far less effort.
According to the report, those behind such operations have used powerful open-source AI models developed in China. Unlike closed systems operated by American companies such as OpenAI, the developer of ChatGPT, or Anthropic, the developer of Claude, these models can be downloaded and modified by anyone.
While major companies can intervene and block users who abuse their closed models, open-source systems can be downloaded directly onto private servers and computing infrastructure. As a result, no technology company retains the ability to prevent or shut down their operation.
Safety Mechanisms Removed
Yedioth Ahronoth said operators exploited these models by stripping away built-in safety mechanisms intended to prevent the creation or manipulation of fake profiles and then connecting them to AI agents.
These agents are autonomous programs capable of operating applications, opening accounts on social networks and performing complex tasks independently.
With the help of such agents, the systems created networks containing thousands of fake profiles and continuously supplied them with posts about political developments and current affairs.
Alongside state-backed campaigns attributed to the Iranian and Chinese governments, security researchers and US intelligence officials also identified two campaigns operated by private commercial companies based in Israel.
The New York Times identified one of the companies as IntelEye, a Tel Aviv-based firm founded in 2023 by former members of NSO Group, the Israeli cyber-offensive company previously blacklisted by the US government.
According to US security officials interviewed about the campaign, IntelEye’s activities attracted the attention of American intelligence companies and social media platforms earlier this year.
The company reportedly used artificial intelligence software built on China’s DeepSeek model to control thousands of social media accounts.
Accounts Both Supporting and Opposing Netanyahu
Around 1,000 of these accounts ultimately became active and participated in heated discussions surrounding Israeli elections.
The fake profiles interacted with social media videos and posts, attempted to engage users and published a range of content, some supporting Prime Minister Benjamin Netanyahu and some opposing him, with the objective of generating engagement and intensifying public debate.
IntelEye co-founder Maor Salik confirmed that the company had used the Chinese model and controlled the accounts, but strongly denied allegations that IntelEye had conducted an influence campaign.
“IntelEye does not conduct influence campaigns,” Salik told The New York Times. “Our work with DeepSeek was conducted solely as part of a research and defensive experiment. Its purpose was to understand how models lacking sufficient safety mechanisms could be abused and to improve the ability to detect such activities in the future.”
Salik acknowledged that the company had acquired around 10,000 social media accounts, approximately 1,000 of which were actively operated by artificial intelligence systems.
He added that the AI model itself selected the topics and wrote the published material without direct human intervention.
In retrospect, Salik said, the company should have informed social media platforms that it was conducting an experiment on their services.
Meta took a more serious view of the operation.
“This appears to be a coordinated attempt to create fake accounts across multiple platforms,” a Meta spokesperson said, adding that the company had removed most of the accounts using its automated verification systems.
A recent Meta report also referred to an “Israel-based commercial operation” using a sophisticated and advanced AI system, although it did not explicitly identify IntelEye by name.
Iranian Campaign Described as “Extremely Concerning”
US officials described the Iranian campaign as “extremely concerning and dangerous”.
According to US intelligence, Iran used exactly the same technology to flood the internet with a network of fake accounts impersonating ordinary American citizens living in major US cities.
The development has increased concern that highly autonomous influence networks could be created without the labour-intensive infrastructure traditionally required for foreign influence operations.
Once deployed, such systems can create personas, publish political content, interact with real users and sustain discussions almost entirely on their own.
AI Use Linked to “Cognitive Surrender”
Separately, a new report from the Massachusetts Institute of Technology has warned that students’ use of artificial intelligence can result in what researchers described as “cognitive surrender”, creating an illusion of learning while isolating students from the learning process itself.
A study involving 26,000 students found that using AI chatbots increased homework scores by 18 per cent but resulted in test scores falling by 20 per cent.
Another MIT study found that students who used artificial intelligence to write research papers forgot the content only minutes later.
MIT researcher Natalia Kosmyna explained that, in such cases, users “did not activate any memory network that would help them remember”.
The findings are now prompting different responses across the United States.
Law schools at Berkeley and Chicago have banned the technology in some courses, while the Los Angeles school district has imposed an almost complete ban on access to AI chatbots.
