Our approach to bioresilience

Biosecurity used to be mostly a public-health problem. Now it's also an AI problem, and Google DeepMind knows it. Together with its drug-discovery sibling Isomorphic Labs, the company has laid out a joint bioresilience strategy that tries to hold two ideas at once: frontier models are powerful enough to help someone build a biological threat, and they're also powerful enough to help the world spot and stop one faster than ever before.

The headline number is partnerships, not products. Over the past year the two organizations say they've built more than 15 working relationships with government bodies, biosecurity groups and research institutions, all aimed at keeping bad actors away from misuse while getting good actors faster access to the tools. It's a quiet kind of infrastructure work — the sort that doesn't generate a flashy demo but matters when an outbreak actually hits.

On the prevention side, the approach centers on a four-step routine applied to models like Gemini: threat modeling, evaluation, mitigation, monitoring. In-house biologists and security staff work alongside outside partners to poke holes in the models before someone else does. There's also a biology-specific spin on SynthID, the watermarking tech DeepMind built for other media, now being adapted so DNA synthesis providers could flag AI-generated sequences that look risky.

Detection gets a computational boost from AlphaEvolve, which DeepMind says can optimize the algorithms behind metagenomic sequencing — the process labs use to sift through genetic material and spot pathogens. Faster, cheaper DNA analysis means outbreak tracking at a scale that wasn't affordable before. The team is also looking at whether AlphaGenome and its Protein Function annotation work could help characterize unfamiliar pathogens straight from sequence data, catching patterns that slower, traditional methods would miss.

And then there's response, where the AlphaFold legacy shows up directly. Trusted researchers are getting access to DeepMind's newer AI systems to speed up vaccine and countermeasure design. Isomorphic Labs, meanwhile, has stood up a dedicated unit that can deploy its drug design engine quickly during an actual outbreak, whether that outbreak is a natural pandemic or something tied to AI misuse. The whole effort sits under DeepMind's broader Frontier Safety Framework, the same playbook it uses for chemical, biological, radiological and nuclear risk more generally.