Google DeepMind and Isomorphic Labs have outlined an AI bioresilience programme built around more than 15 partnerships with government bodies, biosecurity organisations and research groups, AI News reported.

The programme treats frontier AI in biology as a dual-use problem.

The same systems that can accelerate scientific work may also lower knowledge barriers for misuse, so the companies are organising the work around prevention, detection and response.

The named collaborators include CEPI, the Francis Crick Institute, Lawrence Livermore National Laboratory and the UK AI Security Institute.

The network gives the programme a policy and scientific base rather than leaving it as an internal model-safety project.

Prevention covers threat modelling, expert red-teaming and randomised controlled trials to test whether Gemini could help users clear misuse bottlenecks.

The company also uses classifiers, probes and targeted log analysis to detect risky activity.

DeepMind plans to widen the relationships over the next six to twelve months.

It identified threat intelligence, evaluation methods for AI agents and jailbreak mitigations as areas for further work, and is coordinating with the Frontier Model Forum on riskier categories of training data, including virology datasets.

DNA synthesis is one of the concrete risks in the programme.

The company is exploring whether a SynthID-style watermarking approach could help mark biological sequences, but AI News presented the work as exploratory rather than a shipped product.

The longer-term screening goal is functional: predicting whether a novel DNA sequence is likely toxic or pathogenic, not only whether it resembles a known sequence in existing databases.

That distinction matters because synthetic-biology risk can appear in designs that are new enough to evade simple database matching.

Detection work also depends on metagenomic sequencing, which characterises every microorganism in a sample.

A Google and Pacific Biosciences collaboration used the AlphaEvolve coding agent to improve sequencing accuracy, and AlphaGenome is being explored as a way to characterise pathogens directly from sequence data.

The response side centres on medical countermeasures for pathogens that still lack a licensed diagnostic, vaccine or treatment.

DeepMind said more than 10,000 infectious-disease publications have referenced AlphaFold over five years, including work on malaria, dengue fever, Zika, Ebola, Marburg, Lassa fever and Mpox.

Isomorphic Labs gives the programme another route into drug-discovery work, although the current announcement does not turn those research links into a deployed outbreak-response system.

The practical question is how the partnerships move from evaluations, watermarking research and sequencing improvements into standards used by model developers, synthesis providers, laboratories and public-health agencies.