AI used by Israeli intelligence vendor to profile Iranians and Gulf citizens

WASHINGTON (MNTV) — A commercial surveillance platform used the Claude artificial-intelligence chatbot to analyze, classify, and profile the social-media activity of people in Iran and the Persian Gulf — sorting them by political loyalty, location, and demographic group — according to a report from the AI’s maker, Anthropic.

The company said it banned the account in June 2026, after finding the platform was built by, or on behalf of, an entity it named as “S2T Unlocking Cyberspace,” which open-source research suggests is an Israeli-Singaporean commercial intelligence vendor.

The case appears in Anthropic’s Sept. 10 threat-intelligence report cataloging misuse of its models.

As with the report’s other cases, the findings reflect Anthropic’s own investigation; the firm said it caught the platform at its pilot stage, found no evidence it had been used against real targets before the ban, and shared indicators with partners who track surveillance-for-hire firms.

The vendor has not publicly responded, and the account could not be independently verified.

The platform’s core purpose was surveillance, Anthropic said: it mapped the locations of social-media users, sorted them into coded demographic categories — urban, clerical, military, youth, diaspora, and rural — and labeled them as pro-government or opponents, then produced Arabic-language intelligence briefings written in the style of official government reports, complete with sentiment scores by Gulf nationality and recommended “counter-narratives.”

The operator fed Claude batches of social-media posts and directed it to return each poster’s likely demographic, location, and political leaning with confidence ratings, according to the report.

It was also building a stock of more than 255 fake social-media accounts, and had Claude generate posts for personas in Persian, Arabic, English, and German designed to pass as members of both pro- and anti-government populations — suggesting an intent to infiltrate online communities on both sides.

Anthropic said the activity violated its rules against profiling and building dossiers on activists, journalists, and dissidents. The findings, it noted, independently echoed a 2023 investigation by the journalism network Forbidden Stories, which documented an S2T surveillance product in leaked files — a system described as capable of creating fake accounts to infiltrate private WhatsApp and Telegram groups, harvest member lists, and escalate to phishing and device compromise.

Anthropic’s head of threat intelligence, Jacob Klein, said AI was not changing whom such operators target but was making surveillance “cheaper and more efficient.”

Rights advocates have long warned that surveillance-for-hire firms serving Gulf and other governments are used against dissidents, exiles, and minorities — and the report underscores how commercial AI can now automate the profiling of entire diaspora communities, including Iranians and Gulf citizens abroad who have historically been surveillance targets.

The company cautioned that banning an account does not necessarily halt the underlying operation.