Miami
Anthropic says Claude was misused in military, surveillance and repression cases
A new company threat report outlined five cases of abuse involving its AI model over the past eight months. According to Axios, the episodes ranged from missile support and dissident tracking to risky virus-related research requests.
Anthropic said it spent the past eight months tracking and disrupting attempts to misuse Claude in activities tied to warfare, espionage, surveillance and repression, Axios reported, citing the company’s latest threat report.
Among the most serious examples, the company described an Iran-linked operation that used Claude to assemble targeting handbooks focused on U.S. naval forces, including ship locations, personnel, aircraft, ship identifiers, satellite imagery and websites that reveal naval movements. Axios added that other Iran-linked efforts also turned to the model for propaganda, domestic surveillance and the targeting of opposition figures and minorities.
Anthropic also said a weapons cell in northern Yemen relied on Claude Code to develop guidance software for rockets and missiles. According to Axios, the group used separate instances of the model to write code, conduct research and review one another’s work, then returned to Claude within hours of an apparent guided-rocket test failure to troubleshoot what went wrong.
Another case, which Anthropic linked to a China-connected actor, involved searching through more than 100 WhatsApp groups and dozens of Telegram channels to identify Uyghurs in Syria who could be pressured or paid to report on armed Uyghur groups. Axios said Claude also helped a non-Arabic-speaking operator conduct covert outreach in Syrian Arabic by translating replies and advising on attempts to exploit financial hardship, family separation and relatives still in Xinjiang.
The report also included the case of a consultant in Mali who, according to Anthropic, used Claude as the main engineering support for a nationwide surveillance system capable of collecting call records, text messages and voice traffic across 25 million phones. Axios reported that the system could identify people by voice across different SIM cards, flag VPN users and generate intelligence dossiers on any phone number without a warrant.
The company further said it blocked sensitive requests from researchers working under a state-backed grant who were trying to alter chikungunya so it would spread more easily or evade immune defenses, for work intended for a military research institute. Because Claude refused the most sensitive prompts, the platform routed them to a rival model with weaker safeguards, according to the report cited by Axios.
Axios said Anthropic presented the findings as evidence that frontier AI labs are taking on a role similar to intelligence agencies, because their systems can both empower hostile actors and provide early warning of emerging threats. In Washington, the outlet added, AI safety has become a more urgent political issue: a bipartisan group of House lawmakers is urging Mike Johnson to cancel recess until Congress acts, while Bernie Sanders wants to ban superintelligence.
Sources
About this piece. Written by El Notero with AI assistance, from reporting published by the outlet cited above. El Notero did no original reporting on this story.
