Sept. 10 (Reuters) – Anthropic thwarted attempts to use its artificial intelligence models Claude for the development of biological weapons and halted actions linked to a cyber espionage campaign allegedly tied to Russia against Ukraine, the AI giant said in a report published on Thursday.
The company also accused Chinese rivals of intrusion attempts aimed at extracting Claude’s resources. The data were presented in the Threat Intelligence report, which documented the malicious use of its technology over the last eight months.
A growing number of illegal intrusion attempts by AI agents, coupled with industry experts’ concerns about the risks of the technology, has increased pressure on the leading pioneers in the field, raising the prospect of more regulation.
Two Anthropic researchers warned this week that the rapid advancement of artificial intelligence could lead to the extinction of the human race in a not-too-distant future.
MISUSES, DRONES AND BIOLOGICAL WEAPONS
In the report, Anthropic said that there has long been concern that AI models could someday reach a level of capability that would enable them to make existing pathogens more dangerous or to create wholly new ones.
The company documented five examples of scientists using its models in ways that could support the development of biological weapons.
In one case, a researcher from a region not served by Anthropic used a private virtual server infrastructure to access Claude and, for weeks, used it to plan experiments on adapting avian influenza to mammals. Regions not served by Anthropic include countries such as Russia, China and North Korea, among others.
After detecting the misuse, the company banned the accounts involved in these researches and incorporated its findings into the protection, monitoring and threat intelligence processes of its leading models, the American company said.
Anthropic did not identify the institutions, the countries where they are based, or the specific biological agents and research techniques involved.
The company also identified what it called “new categories of threat agents.” This includes the use of the platform to “develop software for conventional weapons, including firearms, missiles, armed drones, bombs and other munitions, as well as the targeting and control systems that operate them.”
The report details incidents involving operators in China, Russia and Yemen who used Claude to develop software for weapon design and development, or to support intelligence gathering and acquisitions related to arms programs.
The rapid improvements in Anthropic’s models have generated new risks, said Jacob Klein, the company’s head of threat intelligence, to Reuters.
“A year ago, say you wanted to optimize a drone or the software of a missile; the models simply wouldn’t be as good at that task as they are now,” he said.
RUSSIAN HACKERS
Anthropic found that criminals and government-backed hackers are increasingly using AI to orchestrate and carry out much of their attacks, with humans often acting as supervisors, rather than direct operators.
“The use of AI went beyond simple Q&A chatbot interactions, actually involving the use of multi-agent frameworks,” the company said.
A group of hackers allegedly conducted phishing operations, hotel Wi‑Fi network takeovers and WhatsApp account intrusions targeting Ukraine’s governmental, military and diplomatic sectors, using AI at nearly every step, Anthropic said.
The group’s technique was consistent with that of the threat actor Midnight Blizzard, based in Russia, which the U.S. government has previously linked to the Russian foreign intelligence service SVR.
The Russian embassy in Washington did not comment on the matter.
The group would have used AI to create a system that automatically detects when its malware is discovered by security defenses and rewrites its programming until it escapes detection.
Anthropic also said it had detected and halted activities tied to affiliates of the ShinyHunters collective, currently one of the most prolific cybercrime organizations, associated with attacks against major corporations worldwide.
CHINESE LABORATORIES
Anthropic said it had interrupted attacks from seven laboratories based in China, including Alibaba, Moonshot, DeepSeek and Xiaomi, during the period covered by the report.
These companies did not immediately respond to a request for comment.
Operators tied to Alibaba conducted what Anthropic called the largest “illicit distillation” attack, allegedly aimed at extracting Claude’s capabilities and using them to enhance Alibaba’s Chinese tech company Xiaomi’s Qwen models.
Anthropic said it observed more than 151 million exchanges attributed to Alibaba between May and July 2026, with a peak of nearly 3 million per day from more than 3,500 accounts it described as fraudulent.
Distillation refers to the process of training smaller AI models using the results of larger, more expensive models, in an attempt to reduce the training costs of a new AI tool.
In another case of misuse of its technology, Anthropic said Moonshot, creator of the Kimi chatbot, and DeepSeek allegedly redirected live customer conversations — sometimes including confidential information — through Claude and used their responses as training data.
China’s Ministry of Foreign Affairs said it was not aware of the Anthropic report and that the government maintains that AI should be developed for the common good and opposes fact distortion and defamatory campaigns against the country.
