DeepMind and Isomorphic Labs set out an AI plan for biosecurity

Google DeepMind and Isomorphic Labs have laid out a shared approach to what they call bioresilience: a society's ability to prepare for and respond to biological threats, whether a natural outbreak or the misuse of advanced AI. The write-up is a strategy document rather than a results paper, but it names specific tools rather than staying abstract. The framing is deliberately two-sided, stopping bad actors from misusing the models while helping governments, scientists, and biosecurity experts use the same systems to build defenses.

The plan splits into three parts. For prevention, the companies are adapting their SynthID watermarking so that DNA synthesis providers can screen orders for risky AI-generated sequences before they are made, alongside threat modeling and evaluations of their own models. For detection, they point to AlphaEvolve being used to improve metagenomic sequencing algorithms, which they say makes tracking disease faster and cheaper. For response, trusted researchers get early access to their systems for vaccine design, and Isomorphic Labs has set up a dedicated unit to aim its drug design engine at medical countermeasures. DeepMind says it has built more than 15 partnerships with government bodies, biosecurity organizations, and research groups over the past year, though it does not yet report outcomes from them.

Why it matters

If you work in DNA synthesis or biosecurity, the concrete piece here is sequence screening built on SynthID, which turns AI-misuse policy into something a synthesis provider can actually run at order time. The rest is a template other labs will be measured against, so it is worth watching whether the 15 partnerships produce results or stay announcements.