(José Niño, Headline USA) An Iran-linked operation used Anthropic’s Claude model to collect information that could help target American naval forces in the Middle East, according to The Wall Street Journal. Anthropic said the group combined transponder data, military photographs, public websites, and commercial satellite imagery to monitor ships and investigate weaknesses in shipboard communications. The company detected the activity and disabled the accounts, though it could not determine whether anyone used the material in an actual military operation.
The case appeared in Anthropic’s 154-page threat report, which described alleged misuse connected to weapons research, surveillance, propaganda, and biological threats. The Wall Street Journal reported that users often concealed their goals or changed tactics after Claude refused prohibited requests. Anthropic bars customers from using its systems for violence, weapons construction, or abusive surveillance.
Anthropic also identified and disrupted attempts to obtain help developing biological weapons, although it did not name the suspected countries. “Without the correct safeguards, such capabilities could have catastrophic consequences,” the report warned.
In northern Yemen, another group allegedly asked Claude for assistance with guidance software for advanced weapons, including a multistage ballistic missile. “Our safeguards blocked many of their requests, but not all of them,” Anthropic said. The company banned the associated accounts, alerted government partners, and introduced additional detection measures.
The Wall Street Journal also cited an operation aligned with the Chinese government that allegedly monitored Uyghurs, including people who had entered the newly created Syrian Army. Anthropic said the same network targeted Uyghur journalists through coordinated reporting and bot amplification. Other China-linked accounts reportedly assembled profiles of Catholic cardinals, Taiwanese Presbyterian leaders, Tibetan Buddhist organizations, and members of the Tibetan administration in exile.
Russia based users pursued different objectives. According to the report, they tried to apply battlefield data from Ukraine to autonomous drone swarms and produce Russian aligned propaganda for audiences in the Central African Republic and Democratic Republic of Congo.
The disclosures arrived amid growing concern inside the artificial intelligence industry. Anthropic researcher Evan Hubinger wrote, “We really do earnestly believe AI could kill all humans!” He continued, “I personally think it is >10% within the next decade. I believe Anthropic is trying its best but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to,” while later clarifying that he considered the danger from present models low.
Anthropic said publication could help others respond more effectively. “We hope that the findings in this report will help other developers recognize similar patterns on their own platforms, give governments and civil society a clearer view of how emerging threats take shape, and strengthen collective defenses,” the company wrote.
José Niño is the deputy editor of Headline USA. Follow him at x.com/JoseAlNino
