Google DeepMind and Isomorphic Labs have launched a bioresilience programme designed to prevent artificial intelligence from being misused in biology while applying the same technology to detect outbreaks and accelerate vaccines and treatments. The initiative reflects growing concern that frontier AI could both increase biological risks and provide essential tools for managing them.
A programme built on three pillars
The joint programme focuses on prevention, detection and response. During the past year, the organisations have developed more than 15 partnerships with government agencies, biosecurity organisations and research institutions.
Named collaborators include Lawrence Livermore National Laboratory, the UK AI Security Institute, the Coalition for Epidemic Preparedness Innovations and the Francis Crick Institute.
Google DeepMind plans to make selected models and AI agents available to trusted researchers. The objective is to improve disease surveillance, identify emerging pathogens and reduce the time required to develop medical countermeasures.
The programme builds upon technologies including AlphaFold, which mapped the three-dimensional structures of almost all known proteins, and AlphaGenome, which predicts how DNA mutations affect gene activity.
Preventing biological misuse
The central challenge is that an AI system capable of assisting vaccine development may also provide knowledge useful to someone seeking to create a biological threat.
DeepMind applies a four-stage safety process involving threat modelling, evaluations, mitigations and monitoring. Experts conduct red-team exercises to test whether Gemini and specialised scientific tools could help users overcome practical barriers to harmful biological activity.
Models are trained to reject dangerous requests while continuing to answer legitimate scientific questions. Automated classifiers, monitoring tools and analysis of usage logs are intended to identify suspicious behaviour.
The company acknowledges that these safeguards remain incomplete. Filters designed around known attack methods may not detect new attempts to bypass restrictions.
DNA screening faces an AI challenge
One area of particular concern is commercial DNA synthesis. Providers traditionally compare customer orders against databases containing known dangerous pathogens and toxins.
AI could potentially design a sequence that performs a harmful biological function without closely matching anything in existing databases. Such a sequence might therefore avoid conventional screening systems.
DeepMind is exploring whether its SynthID watermarking technology could be adapted to identify AI-generated biological sequences. Longer-term research aims to predict whether a novel sequence could be toxic or pathogenic based on its function rather than its similarity to known organisms.
Both ideas remain experimental rather than operational products.
Earlier detection and faster treatment
The detection strategy centres on metagenomic sequencing, which identifies all microorganisms within a sample instead of testing only for selected pathogens. Wider deployment could create early-warning networks using wastewater, hospitals and transport hubs.
DeepMind’s AlphaEvolve agent has already been used to improve sequencing algorithms, while AlphaGenome may help researchers characterise unfamiliar pathogens. Cost remains a major obstacle, particularly in lower-income regions where outbreaks may initially emerge.
For response, Isomorphic Labs has established a dedicated unit capable of deploying its AI-powered drug-design engine during a new outbreak. Trusted researchers will also receive access to systems intended to accelerate vaccines, antibodies and other treatments.
The programme demonstrates the central contradiction of advanced biological AI: greater scientific capability creates both protection and risk. Its success will depend not only on technical safeguards but also on legislation, international coordination and whether early research systems can operate reliably during a real public-health emergency.
Newshub Editorial in Technology – 19 July 2026
Ask NF GPT
If you have an account with ChatGPT you get deeper explanations,
background and context related to what you are reading.
Recent Comments