Anthropic Blocks Bioweapons and Cyber Attacks in Ongoing AI Security Drive
Anthropic's AI Security Efforts and Threat Intelligence Findings
By AJ Vicens and Eduardo Baptista
Sept 10 (Reuters) - Anthropic broke up attempts to use its Claude models to develop biological weapons and carry out a suspected Russia-linked cyber espionage campaign against Ukraine, the AI heavyweight said in a report published on Thursday.
It also accused Chinese competitors of hacking attempts aimed at extracting Claude's capabilities in its latest Threat Intelligence report, which documented malicious use of its technology over the past eight months.
A growing number of illegal hacking attempts by rogue AI agents along with concerns over the technology's risks voiced by industry insiders have increased the pressure on leading pioneers in the field, raising the prospect of more regulation.
Two Anthropic researchers warned this week that rapidly progressing artificial intelligence could lead to the extinction of the human race in the not-too-distant future.
Missiles, Drones and Biological Weapons
AI and the Threat of Biological Weapons
In its report, Anthropic said it has long been a concern that AI models might one day reach the level of capability where they could help make existing pathogens more dangerous or create entirely new ones.
It documented five examples of scientists using its models in ways that could support biological weapons development.
In one case, a researcher in a region unsupported by Anthropic employed virtual private server infrastructure to access Claude and for weeks used it to plan avian influenza mammalian-adaptation experiments.
Regions not supported by Anthropic include countries like Russia, China and North Korea among others.
After detecting the misuse, it banned the accounts involved in such research and incorporated its findings into its frontier model safeguards, enforcement, and threat intelligence processes.
Anthropic did not identify the institutions, the countries where they are based, or the specific biological agents and research techniques involved.
AI in Conventional Weapons Development
It also identified what it called "new categories of threat actors" misusing Claude. That included using 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 detailed incidents of operators in China, Russia and Yemen using Claude to develop software for weapons design and development, or to support intelligence gathering and procurement related to weapons programs.
Rapid improvements in Anthropic's models have raised new risks, Jacob Klein, its head of threat intelligence, told Reuters.
"A year ago, let's say you wanted to optimize a drone or optimize the software on a missile, the models just wouldn’t be as good at that task as they are now," he said.
Russian Hackers Use Anthropic AI Against Ukraine Officials
AI-Driven Cyberattacks and Espionage
Anthropic found that cybercriminals and state-backed hackers were increasingly using AI to orchestrate and execute large portions of cyberattacks, with humans often serving as overseers rather than hands-on operators.
"The use of AI went beyond simple questions and responses from a chatbot but rather involved the use of multi-agent frameworks," it said.
One hacking group allegedly ran phishing, hotel Wi-Fi hijacking and WhatsApp-takeover operations against targets in the Ukrainian government, military and diplomatic sectors, using AI at nearly every stage, Anthropic said.
The group's tradecraft was consistent with Russia-based threat actor Midnight Blizzard, which the U.S. government has previously linked to Russia's SVR foreign intelligence service.
The Russian Embassy in Washington did not immediately respond to a request for comment.
Automated Malware Evasion
The group allegedly used AI to build a system that automatically detected when its malware was flagged by security defenses and rewrote the code until it evaded detection.
Disruption of Cybercrime Enterprises
Anthropic said it also detected and disrupted activity linked to affiliates of the ShinyHunters collective, currently among the most prolific cybercrime enterprises, which is linked to attacks on major corporations around the world.
Chinese AI Labs Run Attacks to Extract Claude Capabilities
Illicit Distillation Attacks
Anthropic said it had disrupted attacks from seven China-based labs 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 it linked to Alibaba ran what Anthropic called the largest "illicit distillation" attack, allegedly aimed at extracting the Claude models' capabilities and using them to improve the Chinese tech firm's Qwen models.
Anthropic said it observed more than 151 million exchanges it attributed to Alibaba between May and July 2026, peaking at nearly 3 million per day from more than 3,500 accounts it described as fraudulent.
Understanding Distillation
Distillation refers to the process of training smaller AI models using output from larger, more expensive models in a bid to lower the costs of training a new AI tool.
Misuse of Claude in Customer Data Handling
In another misuse of its technology, Anthropic said Kimi chatbot creator Moonshot and DeepSeek allegedly routed live customer conversations, which sometimes included sensitive information, through Claude and used its responses as training data.
China's foreign ministry said it was not aware of the Anthropic report and that the government maintains that AI should be developed for good and opposes distortion of facts and smears against the country.
(Reporting by Prathik Jayaprakash in Bengaluru, AJ Vicens in Detroit, Karen Freifeld in New York and Eduardo Baptista in Beijing; Editing by Tasim Zahid, David Gregorio and Joe Bavier)
