Anthropic disrupted several cases involving the misuse of its Claude AI models by state-sponsored or state-linked actors, according to the company’s threat-intelligence report. RTE reported that the cases were identified between January and July. They involved actors linked to China, Iran and West Africa.

The reported surveillance operations targeted diaspora and dissident groups. These included pro-democracy figures from Hong Kong, Tibetan and Falun Gong communities in Asia, as well as Iranian minority groups and opponents of the Iranian government living abroad. In one Iranian case, operators used Claude to build a system that could identify people through their social-media accounts. In another case, a contractor working for Mali’s national-security authorities used Claude to develop intelligence-gathering software.

Anthropic’s threat report is the source of these allegations. Available reports do not independently verify the operations or the identities of all those involved.

The company also reported several other forms of misuse. Politico said Anthropic found fraudulent accounts and online services being used by criminal hacking groups, Chinese security agencies, Russia-linked spies and Yemen-based arms manufacturers to access Claude. Anthropic said it disrupted every case detailed in its report. It also notified relevant government and industry partners.

Al Jazeera reported that operators in northern Yemen used separate Claude sessions for tasks linked to missile guidance and flight-control software. Anthropic said its safeguards blocked many requests. However, some requests went through after the operators divided their work across different sessions and hid the intended purpose. The company said it found no evidence that the group deployed a functioning weapon. It did report an apparent unsuccessful test-fire.

Anthropic also identified five cases in 2026 involving scientists in unspecified foreign countries who used its models to research dangerous pathogens. The company said such work could have links to bioweapons activity. NBC News separately reported that researchers connected to state institutions bypassed regional restrictions to use Claude for dual-use biological research.

The company further accused Chinese developers of using Claude’s outputs to improve their own AI models through distillation. The technique involves training another model using outputs from an existing model. Anthropic’s allegation concerns the unauthorised collection of those outputs. RTE reported that Anthropic named Moonshot and DeepSeek in its report.

According to RTE, Moonshot allegedly routed nearly 300,000 customer requests to Anthropic over 10 days through 5,380 fraudulent accounts. Most of the accounts appeared to be located in Singapore and Japan. Some requests reportedly contained sensitive user information. The available report does not include responses from Moonshot or DeepSeek.

The cases also show how users can bypass regional restrictions and AI safety systems through large account networks, task splitting and seemingly harmless requests. Detecting such misuse requires platforms to examine account activity, request patterns, payment details, network signals and outputs across multiple sessions.

Anthropic threat-research head Jacob Klein told Politico, "We're not trying to be hyperbolic here. We just want to present to the world: Here's what the technology can actually be misused for today."