The artificial intelligence company Anthropic released a report on Sept. 11, 2026, detailing what it described as multiple instances of its Claude AI models being used for malicious activities by state-aligned and other groups. The report outlines alleged attempts to use the AI for developing missile software, conducting cyber-espionage, and running covert influence campaigns.
According to the company’s threat report, Anthropic detected and disrupted an effort in northern Yemen to use Claude for developing missile guidance software. The report stated that the operators assigned the AI model various roles to write code for a guided rocket and a long-range ballistic missile. While internal safeguards blocked numerous requests, Anthropic reported that some prompts were successful because the operators divided tasks across separate sessions to obscure their final objective.
Anthropic said it found no evidence that a functional weapon was developed, though it noted the group appeared to have conducted an unsuccessful test. The company stated it banned the associated user accounts and shared information with public and private sector partners.
The report also detailed several state-linked cyber operations. It alleged that a Russian-aligned group, identified with characteristics of APT29, also known as Midnight Blizzard, used Claude to automate cyberattacks against Ukrainian, European, and diplomatic targets, including drone manufacturers.
Separately, Anthropic said it disrupted an operation by university students in Hunan, China, who allegedly used the AI to engineer and manage an offensive program targeting government and corporate networks in the Middle East, Europe, and Southeast Asia. In response to both incidents, the company said it banned the accounts and increased monitoring for similar activities.
Further incidents outlined in the report include the removal of three Iranian state-aligned accounts that were allegedly using Claude for influence and psychological operations. The company linked these activities to Iranian institutions, including one distributing content associated with the Islamic Revolutionary Guards Corps (IRGC).
In another case, the report described a China-aligned group using Claude to recruit Uyghur individuals in Syria. Anthropic stated the group, which had no Arabic speakers, used the AI to draft outreach messages in a regional dialect and translate responses in real time.
The report’s publication follows recent public statements from former Anthropic employees regarding AI safety. Former researcher Jacob Coxon resigned over safety concerns, and another scientist, Evan Hubinger, publicly agreed with his assessment. These events have contributed to calls from some U.S. lawmakers for increased regulation of AI systems. Anthropic stated it is investigating the breaches and has hired an independent firm to review the incidents.
The company’s relationship with the U.S. government has included legal and policy disputes. Earlier this year, the Pentagon designated Anthropic as a supply chain risk, a decision the company challenged. A judge in California later ruled that the Department of Defense acted unlawfully in issuing the designation. Despite the public disagreement, some reports indicate the Pentagon has used Claude models in military missions.
