Anthropic reported that it blocked several attempts by researchers to use its AI models for activities related to the development of biological weapons. The company identified five specific cases where users attempted to bypass safety protocols and obscure their research goals.
In one instance, a researcher from an unsupported region spent several weeks using the Claude model to plan experiments involving avian influenza. Anthropic restricted this activity to its least capable model and subsequently banned the accounts involved in the reported incidents.
The company noted that it could not determine if the researchers intended to cause harm, as information used for bioweapons can also be applied to vaccine development. Anthropic did not disclose the names of the institutions or the countries where these events occurred.
The report follows the resignation of Anthropic safety researcher Jacob Coxon, who cited internal concerns regarding the potential existential risks of AI. Meanwhile, the industry continues to debate the necessity of stricter regulations for AI applications in biological research.
