Google DeepMind expands AI bioresilience program

Google DeepMind and Isomorphic Labs have outlined a bioresilience initiative aimed at reducing the risk of AI misuse in biology while improving outbreak detection and response. The companies said the program has grown over the past year into more than 15 partnerships with government bodies, biosecurity groups, and research organisations. The effort is organised around three goals: preventing harmful use, detecting outbreaks sooner, and helping respond once an incident begins. DeepMind said its work includes threat modelling, redteaming, and tests designed to see whether Gemini or other AI systems could help users overcome barriers to misuse, while also avoiding blocking legitimate scientific research. One area of concern is DNA synthesis screening. DeepMind said existing screening methods may be less effective as AI can help design DNA sequences with similar biological function without matching known harmful sequences. The company said it is exploring adapted watermarking techniques and other methods to better identify risky designs. On detection, DeepMind is looking at metagenomic sequencing and related dataprocessing improvements that could support earlier outbreak warning systems. On response, it highlighted work with partners such as Lawrence Livermore National Laboratory, CEPI, and the Francis Crick Institute, including research on antibody design and other countermeasure development. The companies also said they are engaging with policymakers on biosafety standards, DNA screening, and earlywarning infrastructure.