Russian Intelligence Used Claude AI for Theft and Election Influence

Anthropic published a report on the abuse of its artificial intelligence models in the period from December 2025 to August 2026. As reported by the BBC, one of the main threats was attempts to use the Claude chatbot for scientific research that could lead to the creation of biological weapons.

Users turned to the neural network for help in writing grant applications for the study of the chikungunya virus. Requests concerned the genetic modification of the pathogen to increase its contagiousness and ability to bypass the human immune system.

The New York Times clarifies that the suspicious work was to be carried out at a military research institute. The technology company could not precisely determine the feasibility of such experiments: they could lead both to the creation of a vaccine and to the development of biological weapons. Due to the high risk, the scientists' accounts were blocked. The report also records attempts to use artificial intelligence for developing software for missiles and combat drones by users from Russia, China, and Yemen. After these incidents, Anthropic strengthened restrictions in its latest models, Claude Fable 5 and Mythos.

Automation of Cyberattacks and Espionage

Artificial intelligence significantly lowered the barrier for conducting large-scale hacking operations. CyberScoop reports data on the activity of an attacker under the pseudonym JackPoterz, whose methods coincide with the Russian state-sponsored group Midnight Blizzard. He used Claude to organize an espionage campaign against more than twenty governmental and defense institutions in Ukraine and Europe.

The neural network continuously monitored the reaction of antivirus programs to malicious code, and upon detecting a threat, autonomous agents immediately rewrote the program. In this way, documentation for drone video surveillance systems and hundreds of thousands of national identification records were stolen.

The technology also accelerated the activities of other groups. Chinese students, with the help of a swarm of AI agents, found dozens of zero-day vulnerabilities in network equipment within a month. And the cybercrime syndicate ShinyHunters stole more than 2100 corporate access keys to Microsoft Azure cloud services in just 34 hours.

Generation of Propaganda and Data Theft

The report describes instances of models being used to create networks of fake accounts. As noted by Sky News, Russian state media used Claude to generate propaganda materials, disguised as independent news to influence elections in Moldova. Similar campaigns were also recorded from Iran and Turkey.

A separate area of abuse was the so-called "distillation" — data theft by Chinese AI laboratories. Companies Alibaba, DeepSeek, and Moonshot AI created thousands of fake accounts to generate up to three million daily requests to older versions of Claude. The obtained responses were used to train their own Chinese neural networks.

Some laboratories secretly redirected their users' requests to the American model, presenting its responses as their own. Among the data thus transferred were surveillance camera recordings, requested by representatives of the Chinese army.

Resignation of a Researcher and the Problem of Control

The publication of the abuse report occurred against the backdrop of an internal crisis in the industry. One of the researchers, Jacob Coxon, resigned from Anthropic. He explained his resignation by saying that the competition between Anthropic and OpenAI leads to uncontrolled technological development. He noted that some of his colleagues consider the destruction of humanity by artificial intelligence possible by the end of this decade. Cybersecurity experts emphasize that a situation in which private corporations independently and without democratic oversight define the boundaries of safe artificial intelligence use creates risks for society.

• 1
• 0
• 0
• 0
• 2
• 0
• 1
• 0