Anthropic blocks multiple attempts to use AI for biological weapons research this year

According to the Financial Times, Anthropic has disclosed that it has blocked multiple attempts by scientists to use its AI technology for research that could aid in the development of biological weapons this year. The company listed five cases where researchers bypassed control measures and used deceptive methods to obscure their research intentions. One case involved a researcher from an 'unsupported region' using the Claude model to plan experiments on avian flu over several weeks, with safety filters restricting the research to the weakest-performing model. Anthropic emphasized it cannot determine the intent of the scientists involved, noting that the same information could be used for both harmful and beneficial purposes. Anthropic has banned the accounts mentioned in the report but has not disclosed the names of the institutions or countries involved. Earlier this week, Anthropic's security researcher Jacob Coxon resigned, expressing concerns that AI may pose an existential threat before the end of this decade. The report also highlights growing concerns among AI executives and biosecurity researchers about the need for safety measures and regulation in AI's application to biology, as AI could be exploited by various actors to create biological weapons or design viruses.