Anthropic reports misuse of AI model Claude by Russian and Chinese actors

According to developer Anthropic, artificial intelligence is increasingly being used to control entire cyber attacks. The company reports on a suspected Russian espionage campaign and Chinese AI companies that are said to have misused its model for their own purposes.

The AI ​​company Anthropic says it has stopped several cases of misuse of its AI model Claude in the past eight months. These included a cyber espionage campaign allegedly controlled by Russia and attempts by Chinese AI companies to copy Claude's skills. Cybercriminals and government hackers are increasingly using AI not just to support, but to control and execute entire cyberattacks, the company's report said.

According to the report, Anthropic stopped attacks from seven Chinese labs, including tech giants Alibaba, Moonshot, DeepSeek and Xiaomi. Employees associated with Alibaba are said to have carried out the largest attack of its kind in order to use Claude's skills to improve the company's own Qwen models. Smaller AI models are trained with the results of larger, more expensive models in order to reduce development costs.

Between May and July 2026, Anthropic detected more than 151 million such interactions from over 3,500 accounts classified as fraudulent. The developers of the chatbots Kimi (Moonshot) and DeepSeek, on the other hand, are said to have fed live conversations from customers, some of which contained sensitive information, into Claude and used his answers as training data.

A hacker group whose approach is consistent with the Russian group "Midnight Blizzard" is said to have used AI to carry out attacks on targets in the Ukrainian government, military and diplomatic sectors. The US government has previously linked "Midnight Blizzard" to the Russian foreign intelligence service SVR. The group is said to have developed a system that automatically detects when its malware is detected by security systems and rewrites the code until it remains undetected again. Anthropic also identified other actors who are using Claude to develop software for conventional weapons such as firearms, missiles and armed drones.

Jacob Klein, head of threat analysis at Anthropic, said in an interview that models have become more powerful over the past year, bringing new risks. "A year ago, the models for a task like optimizing a drone or a rocket's software simply wouldn't have been as good as they are today," he explained. Anthropic also uncovered activities attributed to the cybercrime collective ShinyHunters.