Artificial intelligence company Anthropic has blocked multiple attempts by state-backed actors, hackers and rival tech firms to misuse its Claude AI models for dangerous activities including potential bioweapons development and advanced cyberattacks.
In a recent Threat Intelligence report, Anthropic revealed it stopped five attempts to use its AI for researching dangerous pathogens like chikungunya, avian influenza and smallpox. In one instance a military-affiliated researcher sought assistance that raised concerns about creating more dangerous variants of a virus. While identifying these queries does not confirm an imminent attack, Anthropic warned that highly capable AI systems are lowering the barrier for dangerous biological research.
The company also disrupted a major cyber espionage campaign linked to "Midnight Blizzard" a Russian state-backed hacking group. The group used AI to orchestrate phishing attacks and constantly rewrite malware to evade security defences while targeting Ukrainian government and military sectors.
Additionally, Anthropic accused seven Chinese tech firms including Alibaba, Moonshot, and DeepSeek of "illicit distillation." These companies allegedly ran millions of fraudulent queries to extract Claude’s capabilities to train their own AI models. Disturbingly some firms routed live and sensitive customer conversations through Claude to generate training data.
The report also highlighted AI misuse for conventional weapons development, noting queries from Yemen, Russia and China regarding software for missiles, drones and firearms. Furthermore the Anthropic halted activity linked to the notorious ShinyHunters cybercrime syndicate.
As AI models grow more sophisticated Anthropic warned that threat actors are shifting from using chatbots for basic assistance to deploying automated AI agents for complex malicious operations. In response the company has restricted access to its most sensitive models and strengthened its security safeguards.
