Our approach to bioresilience

Details

Google DeepMind and Isomorphic Labs have published their joint approach to bioresilience, detailing a program that has been running for 12 months across three domains: prevention, detection, and response.\n\nOn prevention: the organizations apply a four-step safety process — threat modeling, evaluations, mitigations, monitoring — to biosecurity contexts, working with in-house biologists and external partners to test models like Gemini against potential misuse scenarios. A notable technical development is the adaptation of SynthID watermarking to biology: the technology could allow DNA synthesis providers to screen for potentially risky AI-generated biological sequences, addressing one of the most concrete biosecurity risks emerging from powerful AI models.\n\nOn detection: AlphaEvolve is being used to optimize algorithms for metagenomic sequencing data analysis, making pathogen surveillance faster and cheaper at scale. AlphaGenome and protein structure tools are also being explored for early outbreak identification.\n\nOn response: the program points to AlphaFold (3D protein structure mapping for nearly all known proteins), Isomorphic Labs' AI-powered Drug Design Engine (IsoDDE), and AlphaGenome as tools for proactive defense and accelerated therapeutic discovery. The 15+ partnerships with government bodies and biosecurity organizations represent institutional-scale commitment, not just research prototypes. This is the first comprehensive public bioresilience framework from a frontier AI lab and will likely serve as a reference point for regulatory discussions on AI and biosecurity.