DeepMind wants AI to spot the next pandemic early. The problem is what the same technology enables

A plan to prevent, detect and respond

Google DeepMind and Isomorphic Labs have published a joint bioresilience approach. It aims both to stop malicious use and to give governments, scientists and biosecurity experts useful tools. The companies say they have advanced more than 15 partnerships in the past year.

This is not a machine that predicts the date and location of the next pandemic. It is an attempt to gain time by improving prevention, surfacing biological signals earlier and accelerating diagnosis or treatment once a threat appears.

Where AI could genuinely help

DeepMind says AlphaEvolve can optimise algorithms used in metagenomic analysis. Faster, cheaper analysis could surface unusual patterns earlier and extend surveillance to regions with fewer resources. But an alert still needs samples, hospitals, confirmation laboratories and public authorities able to act.

The same capability can make harm easier

A model that understands proteins, genomes and laboratory protocols may help design a drug, while also compressing specialised knowledge for a malicious user. DeepMind proposes threat evaluations, safeguards and controlled access, and is exploring whether SynthID could help identify AI-generated biological sequences.

Who gets trusted access?

A closed system can obstruct legitimate science and exclude lower-income countries; a fully open one can distribute dangerous capabilities. Trusted-partner status needs external oversight, understandable criteria and a route to appeal—not just the approval of the company operating the model.

The next pandemic will not be prevented by a headline. The test is whether these systems reduce verified response times while remaining accountable to public institutions.