DeepMind, Isomorphic launch bioresilience program

Google DeepMind and Isomorphic Labs launched a bioresilience program that has built more than 15 partnerships with government agencies, biosecurity organizations and research institutions over the past year. The program aims to limit AI misuse in biology and speed detection and response to infectious outbreaks and biological attacks.

The initiative rests on three pillars: preventing misuse, detecting outbreaks earlier and responding to outbreaks or attacks. Named collaborators include Lawrence Livermore National Laboratory, the UK AI Security Institute, CEPI and the Francis Crick Institute. The companies plan to expand partnerships over the next six to 12 months to work on threat intelligence, evaluation methods for AI agents, jailbreak mitigations and handling higher-risk training data such as virology datasets.

Prevention work uses threat modelling to identify likely actors and current bottlenecks, expert red-teaming and randomized controlled trials to test whether a frontier model like Gemini could help close those bottlenecks. Post-training interventions aim to teach models to refuse harmful queries while preserving legitimate scientific access. Real-time classifiers and probes flag risky activity and targeted log analysis seeks subtler misuse that automated filters might miss. The companies describe these as ongoing measures rather than finished solutions and note that performance in controlled tests may not match results against novel attacks in live use.

One risk under study is DNA synthesis screening. Members of the International Gene Synthesis Consortium screen orders against databases of known pathogens and toxins and run screening algorithms. Generative AI can design sequences with similar functions that avoid sequence-level matches and may bypass current filters. The teams are exploring whether techniques related to SynthID watermarking for AI-generated images and text could be adapted to biological sequences. Longer-term work seeks screening methods that predict whether novel sequences are likely to be toxic or pathogenic based on function rather than sequence similarity.

Detection efforts focus on metagenomic sequencing, which reads all genetic material in a sample rather than testing for a short list of known pathogens. Scaling metagenomic surveillance in regions where outbreaks are most likely requires lower sequencing costs. DeepMind points to a Google and Pacific Biosciences collaboration that used an AlphaEvolve coding agent to improve sequencing accuracy and says it is exploring algorithmic and hardware improvements. The group is also investigating whether tools such as AlphaGenome could help characterise pathogens directly from sequence data. These remain research collaborations, not deployed early-warning systems.

On response, the program highlights structure prediction and protein design. AlphaFold has been cited in thousands of infectious disease publications over five years and the update announces a partnership with Lawrence Livermore to use AlphaFold 3 for broad-spectrum antibody design, including a pan-filovirus effort. DeepMind plans to add protein structures and complexes to the AlphaFold Protein Structure Database with priority for countermeasure development. Access to newer agent systems, including Co-Scientist, will be extended to selected researchers, including teams at US Department of Energy national labs under the Genesis Mission. Isomorphic Labs has created a unit to deploy its drug design engine in a novel outbreak and pledged $7 million to Health for Human Potential to support infectious disease research across Asia.

DeepMind and Isomorphic mapped specific policy requests to prevention, detection and response. Their policy priorities include the AI-Ready Bio-Data Standards Act (H.R. 7907), mandatory DNA synthesis screening through the Biosecurity Modernization and Innovation Act (S. 3741), the SCALE Biology Act (H.R. 8981), expanded metagenomic sequencing supported by the America’s Living Library Act (S. 4023), and the Web of Biological Data Act (H.R. 9307 / S. 4770). The companies called for increased DARPA and HHS funding for early-warning research, investments in manufacturing capacity kept ready for rapid activation, pre-established clinical trial networks and faster regulatory pathways. None of the referenced legislation has been enacted.

The program is presented as iterative. The companies expect the program’s effectiveness to be tested over the next six to 12 months as partnerships expand and federal biosecurity policy and funding decisions evolve.