Anthropic has blocked multiple accounts after users tapped its Claude AI models for biological research that could potentially support the development of biological weapons, CNN reported Sept. 10.
The AI company outlined five cases in a Sept. 10 threat intelligence report involving research on viruses, venoms and toxins. Anthropic said users in the cases circumvented controls restricting access from unsupported regions or took other steps to obscure the purpose of their research and evade safeguards.
Anthropic said the individuals implicated in the cases were working scientists, but it withheld their names, research institutions and countries. The company said it does not assert that the researchers intended harm.
The cases included virologists working on a state-sponsored grant involving gain-of-function research on the chikungunya virus. The research focused on the virus’ transmissibility and immune evasion, according to Anthropic.
In another case, Anthropic identified a researcher outside the U.S. using Claude for work involving highly pathogenic avian influenza, or bird flu. The research focused on mammalian adaptation and mechanisms associated with severe illness. Anthropic said Claude’s assistance in that case was primarily limited to areas such as data analysis, study planning and writing.
Other cases involved research on orthopoxviruses and novel venoms and toxins. Anthropic said some of the work had potential therapeutic applications while also carrying the potential for harmful use.
In a separate review of 30 days of activity associated with what Anthropic called adversarial state institutions, the company found roughly 35 distinct research efforts. Most involved ordinary civilian science, but some had notable potential for both beneficial and harmful applications, according to the report.
Anthropic said earlier Claude models were below the threshold at which they could meaningfully assist sophisticated users with dangerous biological research. The company said it can no longer make the same assurance about today’s models as their scientific capabilities advance. Recent models, including Claude Fable 5, have been launched with stronger safeguards restricting access to a range of dual-use biological research requests.
Still, Anthropic cautioned that its findings are not evidence that Claude is currently making biological threats imminent. Instead, the company said the cases demonstrate the challenge of distinguishing legitimate scientific work from potentially harmful research when the same biological knowledge can be used for either purpose.
The disclosure comes amid broader concerns about the safety of increasingly capable AI systems. CNN reported that former Anthropic employee Jacob Coxon, who recently left the company over AI safety concerns, warned this week that more advanced AI systems could eventually cause significant harm, including through attacks on critical infrastructure or biological weapons.
At the Becker's 11th Annual IT + Revenue Cycle Conference: The Future of AI & Digital Health, taking place September 14–17 in Chicago, healthcare executives and digital leaders from across the country will come together to explore how AI, interoperability, cybersecurity, and revenue cycle innovation are transforming care delivery, strengthening financial performance, and driving the next era of digital health. Apply for complimentary registration now.
