Anthropic blocks attempts to use Claude for bioweapons research
Anthropic has published a report describing how it detected and stopped attempts between December 2025 and August 2026 to use its AI model Claude for dangerous purposes.
According to Anthropic, this included five cases in which people tried to use Claude for research related to biological weapons. In one example, someone used an anonymous email address and a VPN to exchange thousands of messages seeking help researching a highly contagious flu virus.
In another case, Anthropic says a scientist tried to use Claude to write a grant application for research on a virus that was meant to take place at a military research facility. The company blocked the access out of caution, even though it remained unclear whether the intent was harmful.
The report also lists attempts to use Claude for developing conventional weapons, including missiles and drones sought by groups in Yemen and anti-torpedo technology linked to China. Anthropic also says the AI was used by groups connected to Russia and Iran for espionage and propaganda.
Anthropic further accuses Chinese firms including Moonshot and Deepseek of secretly routing queries through Claude to train their own AI models. These claims could not be independently verified from the available sources.
Jacob Klein, Anthropic's head of threat intelligence, told the New York Times it was an "incredibly nuanced situation," since the same information that could be used to build a weapon could also be used to develop a vaccine.
The report comes shortly after a former senior safety researcher at Anthropic publicly warned about the risks AI poses to humanity.
WHY IT MATTERS
The report shows that AI models are already being deliberately misused for dangerous purposes like bioweapons and weapons research, even as providers try to prevent it.
WHAT YOU CAN DO NOW
- Stay skeptical of AI answers about dangerous substances If an AI gives you information about chemicals, viruses, or weapons, always check it against trusted official sources. Don't rely on the chatbot alone.
- Report suspicious use of AI tools If you notice someone trying to use an AI tool for dangerous purposes, you can report it to the provider or authorities. Many companies offer specific forms for this.
- Read safety reports from major AI providers Companies like Anthropic regularly publish reports on detected misuse of their AI. Checking these shows you what risks exist and how seriously they are taken.
- Only use AI services through official channels Avoid shady resellers or anonymous access to AI tools, since these are often used to bypass safety rules. Sign up with your real identity at trustworthy providers.