Google DeepMind and Isomorphic Labs Launch Bioresilience Program Against AI Misuse in Biology

Google DeepMind, together with its sister company Isomorphic Labs, has laid out details of a bioresilience program designed to limit the risk of AI models being used to design pathogens, while also speeding up the detection of and response to disease outbreaks. The program has been in development for the past 12 months and already spans more than 15 partnerships with government institutions, biosecurity organizations, and research centers.

Three pillars of the program

The program rests on three areas of work: preventing the misuse of AI models in biology, speeding up the detection of disease outbreaks, and improving the response to health crises. Google DeepMind openly acknowledges that the same biological knowledge that helps a scientist design a vaccine target could, in theory, also help someone with malicious intent fill gaps in their own knowledge of pathogens.

To assess this risk, the company relies on a combination of expert red-teaming and randomized controlled studies. More than 50 specialists in biosecurity, bioethics, and AI system safety were involved in evaluating the capabilities and risks of successive versions of the AlphaFold model. The conclusions from these consultations reportedly indicated that the potential benefits of AlphaFold 3 significantly outweigh the marginal risk tied to biological weapons.

AlphaFold and new partners

The newest addition to the program is a collaboration with Lawrence Livermore National Laboratory's bioresilience program, in which AlphaFold 3 will be used to design broad-spectrum antibodies. One specific goal is developing an antibody effective against the entire filovirus family, which includes the Ebola virus.

DeepMind also plans to keep expanding the AlphaFold Protein Structure Database with additional protein structures and complexes later this year, prioritizing targets relevant to developing countermeasures for disease outbreaks. The database has served for five years as a reference point for infectious disease researchers, cited in more than 10,000 scientific papers.

Access to research agents

The company is also expanding access to its agentic systems that support scientific research, including the Co-Scientist tool and the AlphaGenome model, to a select group of scientists, including researchers at U.S. Department of Energy national laboratories under the Genesis Mission program. Isomorphic Labs additionally pledged $7 million in support to the organization Health for Human Potential.

Gaps in DNA synthesis screening

The article points to the declining effectiveness of existing screening methods used when ordering synthetic DNA. These systems rely on recognizing sequences of known pathogens, yet AI models can design sequences with disease-causing properties that don't match any known pattern, allowing them to slip past safety filters.

One proposed alternative is detecting threats through metagenomic sequencing, meaning analysis of the full genetic material collected from environmental samples. The method remains expensive and difficult to access in resource-limited regions, which limits its practical use as a large-scale early warning system.

DeepMind is also working on adapting its SynthID content-watermarking system to DNA sequences, which would help identify genetic material designed or modified using AI before it reaches a lab fulfilling a synthesis order.

A test for US legislation

Google DeepMind backs six specific bills currently before the US Congress that would address controls on DNA synthesis and oversight of AI-related biological risk. None of these bills has been passed yet, and their fate will determine whether the announced bioresilience program actually translates into firmer regulation or remains a goodwill declaration from the company.

Organizations involved in the program, including CEPI (the Coalition for Epidemic Preparedness Innovations) and the Francis Crick Institute, have announced plans to expand cooperation over the next 6 to 12 months. For Polish research institutions and biotech companies, this means access to some of these tools, including AlphaFold 3 and Co-Scientist, could eventually extend beyond the current circle of US and UK partners, though for now the program remains focused on institutions in the US and UK.

The story feeds into a broader debate over the dual-use nature of AI models in the life sciences, in which companies like Google DeepMind are trying to both promote the research benefits of their tools and demonstrate that they are actively managing the risk of those tools being misused for harmful purposes, including the potential design of biological weapons.