According to the Financial Times, as experts grow increasingly concerned about the threats AI poses to public safety, Anthropic has stated that it has blocked multiple attempts by scientists to use its technology for research that could help in the development of biological weapons this year.
Anthropic listed five cases. In these cases, the actors "bypassed control measures" and used other means to "confuse" their research purposes in an attempt to circumvent safety protocols. In a report on the misuse of its models for malicious activities, Anthropic said: "We hope that by sharing these cases, we can spark discussions within the AI industry and governments around emerging bio-risks and how best to address them."
One Avian Flu Experiment Was Restricted to the Weakest Model
Case studies provided by the company show possible misuse of AI in the field of biology, including a researcher from an "unsupported region" using the Claude model to plan experiments involving avian flu over several weeks. Anthropic stated that its safety filters restricted this research to the weakest-performing model.
Anthropic emphasized that it cannot determine whether the scientists in these cases had harmful intentions—the same information required to develop biological weapons could also be used to develop vaccines. Currently, Anthropic has banned the accounts mentioned in the report, but it did not reveal the names of the institutions involved or the countries where the incidents occurred.
Security Controversies Rise, Experts Call for Regulation in the Biological Field
Earlier this week, a storm of public opinion surrounding AI safety began: Anthropic's security researcher Jacob Coxon announced his resignation, stating that the company's developers "truly believe that AI may kill us all before the end of this decade." Earlier this year, concerns about this technology escalated with the release of advanced models like Anthropic's Mythos; OpenAI also caused panic in July when it stated that its model had independently hacked into the AI company Hugging Face.
As AI models become increasingly advanced, there is growing consensus between AI executives and biosecurity researchers that safety measures and regulation are needed for the application of AI technology in the field of biology. Experts are increasingly worried that AI could be used by terrorist groups, state actors, or lone-wolf attackers to create biological weapons, design viruses, or release existing harmful pathogens. However, even if AI can be used to design theoretical biological weapons, real-world obstacles still exist—such as whether they can actually be manufactured and the resources required to do so.
