A man holds an Iranian flag near an anti-US billboard depicting US President Donald Trump and the Strait of Hormuz, in Tehran, Iran, May 30, 2026. Photo: Majid Asgaripour/WANA (West Asia News Agency) via REUTERS

Iran-linked operatives have used artificial intelligence (AI) to profile hundreds of Israelis and members of the Jewish diaspora as part of a broader surveillance and cyberwarfare campaign, according to a new report from leading AI company Anthropic.

The operatives used Anthropic’s Claude chatbot to automate the collection and analysis of publicly available information about their targets, turning what would normally be a labor-intensive intelligence operation into a largely automated process, the company said. The targets included Israeli government and nongovernmental figures as well as members of Jewish communities outside Israel, Anthropic explained.

Anthropic attributed the operation to an “Iran-nexus threat actor” it designated GTG-30004. According to the report, the perpetrators used Claude to build an automated open-source intelligence system that could take lists of targets and assemble detailed profiles on them.

The findings, outlined in Anthropic’s latest threat-intelligence report, offer an overview of how governments and criminal groups are using AI for cyberattacks, surveillance, and influence operations.

The Israeli and Jewish targeting was only one part of the Iranian activity identified by Anthropic. The same actor also used Claude to assist with malware development, including modifying an open-source credential-stealing tool and developing a custom C++ build and obfuscation system, according to the report.

Anthropic said the use of AI allowed the operators to dramatically accelerate activities that previously would have required teams of human analysts and programmers. Anthropic also identified what it described as Iranian domestic-security organizations using Claude to develop surveillance capabilities targeting Iranian dissidents, including political opponents and members of the Iranian diaspora.

In one case, operators analyzed more than 155,000 posts on X and identified dozens of Iranian opposition and diaspora accounts for further monitoring, according to Anthropic. The company said Iranian operators also developed tools capable of linking phone numbers to identities, collecting social-media information and creating phishing pages.

Anthropic claims it identified 16 Claude accounts operated by two linked units associated with Iranian paramilitary and domestic-security agencies. The company further identified Iranian state-linked influence operations that used Claude to produce propaganda, create fake personas, and make government messaging appear to come from independent journalists or ordinary users.

The report highlights a growing concern for national-security officials that AI can make sophisticated intelligence and cyber operations considerably cheaper and faster.

Rather than manually researching hundreds of people, an operator can use an AI system to automate searches, organize information, and generate reports. Anthropic said that was essentially what occurred in the operation targeting Israelis and members of the Jewish diaspora.

The company emphasized that Claude was not independently selecting the targets or carrying out physical attacks. Human operators directed the system and incorporated its capabilities into their existing operations. The Iranian actors also used Claude for military reconnaissance, including gathering publicly available information about US naval forces, according to Anthropic.

Anthropic said it shut down the accounts involved in the operations and developed new detection mechanisms based on what it learned from the cases. The company also said it shared information about the activity with government authorities and other organizations.

The report comes as AI companies increasingly confront the possibility that their models can be used by intelligence agencies, militaries, criminal groups, and political organizations to perform work that once required specialized human personnel.

Anthropic CEO Dario Amodei released a statement arguing for the slowdown of AI development, warning that the rise of the technology could pose unpredictable and existential dangers to the public.

Over the past week, several high-level former employees of Anthropic and Open AI have penned statements cautioning that humans could lose control of the potentially lethal technology.