Anthropic, faithful to its policy of apparent transparency, has created the most detailed catalog ever made about the malicious use of its artificial intelligence Claude. A 154-page report details numerous threats that the company led by Dario Amodei has managed to neutralize. Or so it believes. “The cases we share here are not typical examples of misuse, but rather examples of the most notable and novel threat activities we have identified to date,” the text warns.
Read more Vorágine
Just 48 hours after an Anthropic researcher, Jacob Coxon, announced he was resigning from the company because he believed this technology has the potential to end the human species, the company has shown the catalog of everything it has had to stop to prevent that prophecy from becoming reality.
The biological threat is one of the most concerning in that apocalyptic catalog. “Misuse in the biological field is one of the most serious risks of cutting-edge AI models,” Anthropic acknowledges. It also admits that, “without proper safety measures, such capabilities could have catastrophic consequences.” The report details threats detected between December 2025 and August 2026 in seven risk areas: cyber operations, influence, surveillance, scams and fraud, misuse for biological purposes, conventional weapons development, and AI model distillation.
The report details five cases of possible malicious uses in the biological field that were detected, so the accounts driving those dangerous works were blocked.
One of the most disturbing cases in this field came from an account using Claude from a scientific perspective to obtain a state grant for research on the chikungunya virus, which is transmitted by mosquitoes.
According to Anthropic, which did not identify the country of origin of those queries to Claude, this research could serve equally for vaccine development or for strengthening the virus in a way that would cause greater harm. Anthropic explained to The New York Times that this case caused them special concern because the scientific study came from a military research institute.
Groups related to Russian espionage attacked official bodies in Ukraine and Europe
Those who used Anthropic’s AI to develop software intended for the design and development of weapons, or to gather information and acquire material for armament programs, came from China, Russia, and Yemen.
Read more The crisis claims the first resignation in Ceuta over the CNI warnings narrative
The document points out that “among the malicious actors analyzed are groups allegedly sponsored by states, criminals with economic motivations, commercial spyware providers, state propaganda institutions, and individuals with political motivations.”
The range of cases in which artificial intelligence can be used to cause harm spans “from a network of fake dating apps designed to scam users to surveillance systems created to identify and control dissidents.”
The report suggests that recent AI advances are drastically reducing the capability gap between state-backed groups, which have large resources, and individual attackers. This technology now develops the entire offensive chain.
The company assures that without safety measures, the consequences “could be catastrophic”
A group apparently related to Russian state espionage attacked government, military, diplomatic, and defense bodies in Ukraine and Europe, as well as targets linked to U.S. foreign policy and drone technology. Anthropic detected that the targets were ministries, intelligence services, embassies, research centers, and defense sector companies. The main focus was Ukraine.
Another part of the report states that seven Chinese AI labs including Alibaba, Moonshot, DeepSeek, and Xiaomi extracted or resold Claude’s capabilities to their clients.
Anthropic attributed a distillation attack to operators linked to Alibaba, a system by which a new AI model extracts the capabilities of a superior already trained model in order to reduce the costs of these works.
Read more The Ceuta migration crisis sinks Spanish tourism to Morocco
