Today we’re sharing some case studies of threat actors attempting to use Claude for malicious activity, including developing biological weapons. This is the first time any private company, to our knowledge, has publicly shared evidence of potential misuse of their platform for biological weapons development. Our goals are to enable the professional, legitimate life sciences research done with Claude, realize the incredible beneficial impacts for human health and medicine, and ensure that every AI platform is able to recognize and prevent misuse. Sophisticated, state-affiliated actors don’t tell Claude “I’m from adversarial nation X, build me a bioweapon.” They deliberately mask their origin and go to extreme lengths to obfuscate their intent and make their requests seem benign. When we detect and disrupt them, they route traffic to models with more permissive safeguards. Some attempts are easier to detect, like attempts to engineer viruses to infect humans or spread more efficiently. Our Life Sciences Verification Program is designed to enable beneficial research and prevent harmful use cases. We know our bio safeguards are not perfect, and can be a source of major frustration for professional researchers, but we are iterating constantly to make them better at preventing misuse and to improve the user experience for legitimate research. https://www.anthropic.com/threat-intelligence-report-september-2026

We are sleepwalking into a potentially huge disaster And it amazes me how little attention it is getting.

Expanding access to HIV self-testing through the distribution of HIV self-testing kits and the hiring of peer-based support workers: https://ow.ly/22N7106EwAj

📢 New recommendations by the NHS doctor-led National AI Commission, informed by 12,000+ people's views, on how the UK can safely speed up access to AI to support clinicians and benefit patients balancing innovation with patient safety. Read more 🔗https://brnw.ch/21x5zQN