San francisco: US artificial intelligence firm Anthropic has released a comprehensive threat intelligence report that highlights numerous state-linked cyber intrusions, automated weapons design efforts, and covert influence operations utilizing its flagship model, Claude.

According to Anadolu Agency, the report from the San Francisco-based company details how its Threat Intelligence team identified and disrupted operations involving unauthorized actors attempting to exploit Claude for malicious activities between December and August. The report categorizes these activities into seven primary harm areas: cyber operations, foreign influence campaigns, surveillance architectures, fraud, biological misuse, conventional arms development, and model distillation, which involves unauthorized extraction of an AI system's reasoning by rival developers.

Anthropic claims to have dismantled offending accounts, enhanced internal safety filters, and shared intelligence with government authorities and industry peers where necessary. The report specifically highlights cyber espionage efforts by foreign intelligence services, including a Russian state-linked actor known as Midnight Blizzard, which allegedly used Claude to target military and diplomatic networks in Ukraine, Europe, and the US.

The report further alleges that Chinese-speaking operators linked to security research firms and computer engineering students in Hunan province created an autonomous "exploit foundry," identifying zero-day vulnerabilities in commercial network appliances without human intervention. The report also documents instances of threat actors applying Claude's coding and reasoning tools to kinetic weapons programs, including missile engineering efforts in Yemen.

The report claims that state-backed influence networks in Russia, Iran, and the United Arab Emirates integrated Claude into foreign propaganda pipelines, with Russian operatives allegedly using it to script pro-Kremlin broadcasts and forge documents. Additionally, the report highlights the use of Claude by Iranian propaganda bodies to generate misleading statements and compile profiling databases against political dissidents.

Anthropic's report also reveals the design of a mass surveillance platform in Mali and the exploitation of Claude for dual-use biological research, including gain-of-function research on viruses. The report accuses Chinese AI laboratories of conducting industrial-scale distillation attacks to harvest Claude's reasoning for training their architectures.

While detailing extensive foreign exploitation, the report omits examination of the US military's integration of Claude onto classified defense networks during recent military engagements. Despite a $200 million contract with the Pentagon, a legal clash ensued over restrictions on the use of Claude, leading to the Pentagon's eventual exclusion of Anthropic from new defense contracts.

Anthropic's report does not provide data on whether its models or competing commercial systems were utilized in support of US military operations, leaving the full scope of military AI deployment undisclosed.