Our approach to bioresilience
Google DeepMind and Isomorphic Labs describe their joint bioresilience program, organized around preventing misuse of models like Gemini, detecting outbreaks earlier, and accelerating response. Concrete efforts named include a four-step safety process (threat modeling, evaluations, mitigations, monitoring) under DeepMind's Frontier Safety Framework, adapting SynthID watermarking to flag AI-generated biological sequences for DNA synthesis screening, using the AlphaEvolve agent to optimize metagenomic sequencing algorithms for cheaper pathogen surveillance, and applying AlphaGenome and Protein Function annotation to pathogen characterization, alongside Isomorphic Labs' Drug Design Engine (IsoDDE) for rapid countermeasure design.
What it claims
Google DeepMind and Isomorphic Labs say they have advanced more than 15 partnerships over the past 12 months with government bodies, biosecurity organizations and research groups. Those partnerships span prevention, detection and response. The page is a program announcement, not a release. It contains no model, benchmark, dataset, API or code.
How to use it
None of the named systems is offered for general use. The page says the AI models and agents are being made available to trusted partners only:
- Response work: "trusted researchers" are being granted access to DeepMind's latest AI systems for vaccine and countermeasure design.
- Isomorphic Labs' Drug Design Engine (IsoDDE): a dedicated unit will deploy it for government bodies and non-profit organizations during novel outbreaks. It covers both natural pandemics and risks from AI misuse.
- Getting access: the route is the partnership call linked in a fuller update. The page does not state the eligibility criteria, terms or licence.
Limits
Several items the summary lists as concrete efforts are, on the page, still in progress or exploratory:
- SynthID for biology is described as "working on adapting". The page says it "could help" DNA synthesis providers screen AI-generated sequences. No deployment, accuracy or provider integration is reported.
- AlphaGenome and Protein Function annotation for pathogen detection and characterization are only being "explored". There are no results.
- AlphaEvolve is said to "optimize algorithms" for metagenomic sequencing. The page claims faster, cheaper and more accurate DNA analysis, but gives no figures, baselines or named pipelines.
- The four-step safety process (threat modeling, evaluations, mitigations, monitoring) is named but not detailed. The page places it within DeepMind's broader Chemical, Biological, Radiological and Nuclear (CBRN) risk approach and the Frontier Safety Framework, and gives no evaluation results.
The authors call the program a "complex, long-term effort". An engineer looking for a screening tool, a surveillance algorithm or model access should treat this as a pointer to a partnership channel, not as something adoptable today.
Organisations named here
- Google DeepMind
- Isomorphic Labs
- AlphaEvolve
Technologies named here
- Gemini
- SynthID
- AlphaFold
- IsoDDE
- AlphaGenome
- AlphaEvolve
- Protein Function annotation
- Frontier Safety Framework
Techniques named here
- Four-step safety process
Related in ai-safety

