Anthropic reveals how criminals are using Claude for attacks

Anthropic says its Claude AI models have been used in attempts to support biological weapons research, attack Ukrainian officials and extract the capabilities of its AI systems, showing how quickly advanced models are moving from simple assistance to active roles in malicious operations.

The findings come from Anthropic’s latest Threat Intelligence report, which covers malicious activity detected over the past eight months. The company said it disrupted five cases involving biological weapons research, along with cyber operations and attacks allegedly linked to Chinese AI companies.

According to Anthropic, the researchers made use of Claude in a way that would facilitate biological warfare weapon development, which included the use of a researcher in an unsupported region to access Claude through virtual private servers.

This particular researcher was allegedly accused of using Claude for weeks to conduct experiments regarding adapting avian flu to mammals. The company never revealed the researcher, the institution, or the country.

Furthermore, the firm claimed to have blocked the accounts associated with the activities and applied them to improve its safeguards and threat intelligence system.

Moreover, Claude was also accused of being used to develop software for conventional weapons like missiles, armed drones, bombs, and target systems.

Anthropic also detected a cyberattack against Ukrainian government, military, and diplomatic personnel. Anthropic said the group employed Claude during their operations, where the methods include phishing, hacking into hotel Wi-Fi networks, and compromising WhatsApp accounts.

Anthropic said the activity was similar to the methods used by Midnight Blizzard, a Russian-linked group that the US government said was connected to the SVR intelligence service.

The attackers reportedly made use of AI technology to determine when malware was blocked and automatically modify the code to evade detection.

Anthropic said seven China-based labs, including Alibaba, Moonshot, DeepSeek and Xiaomi, were involved in attempts to extract Claude's capabilities.

Alibaba allegedly carried out the largest illicit distillation campaign, generating more than 151 million exchanges between May and July 2026 across more than 3,500 fraudulent accounts.

In a separate case, Anthropic accused Moonshot and DeepSeek of routing live customer conversations through Claude and using its responses as training data.