Anthropic, OpenAI and Google DeepMind expand biological-risk tests and access controls

Anthropic, OpenAI and Google DeepMind are expanding biological-risk testing, access controls and other safeguards on frontier models they also hope will speed drug discovery, aid disease research and help prevent pandemics. The three labs are treating biology as a dual-use problem: the same systems that can design proteins, troubleshoot lab work and scan the literature can, in the wrong hands, lower the expertise needed to plan a biological weapon. Executives and biosecurity specialists have warned that more capable models could help users create novel viruses or toxins even as the companies sell those capabilities into pharmaceutical research.