The revelation last month that Claude was used "in ways that could support biological weapons development" was made possible, in part, because Anthropic's AI models operate on a closed system overseen by the company.
Read the original report for the complete account of this story.