Google DeepMind: AI Bioresilience Program Aims to Prevent Biological Misuse

Google DeepMind and Isomorphic Labs have expanded a joint bioresilience program designed to reduce the risk of artificial intelligence misuse in biology while accelerating the detection and response to infectious disease outbreaks.

The organisations revealed that the initiative, which began quietly, has established more than 15 partnerships with government agencies, biosecurity organisations and scientific research institutions over the past year.

As AI systems become increasingly capable of analysing complex biological data, DeepMind acknowledged that advanced models such as Gemini, together with specialised biology platforms and scientific databases, can significantly accelerate medical research. At the same time, the company warned that the same capabilities could help malicious actors gain access to sensitive biological knowledge.

To address that challenge, the program focuses on three priorities: preventing AI misuse, improving outbreak detection and strengthening responses to biological emergencies.

Among the initiative’s collaborators are Lawrence Livermore National Laboratory, the UK AI Security Institute, the Coalition for Epidemic Preparedness Innovations (CEPI) and the Francis Crick Institute.

Kimi K3: Moonshot Unveils World’s Largest Open-Weight AI Model

DeepMind plans to expand these partnerships over the next six to 12 months, with additional work focused on threat intelligence, AI agent evaluation, stronger jailbreak protections and safer handling of high-risk biological training data.

Strengthening AI Safety

DeepMind uses threat modelling to identify how individuals could misuse advanced AI systems and to understand the barriers that currently prevent biological threats.

The company combines expert red-team testing with controlled research studies to evaluate whether Gemini could help users bypass those safeguards.

Engineers have also developed post-training techniques that encourage the model to refuse dangerous biological requests while continuing to answer legitimate scientific questions. DeepMind supplements those safeguards with real-time classifiers, behavioural monitoring and targeted log analysis to detect suspicious activity.

The company emphasised that these protections continue to evolve and acknowledged that no safeguard can completely prevent new forms of AI misuse.

Addressing DNA Synthesis Risks

One of the program’s primary research areas focuses on DNA synthesis screening.

Current DNA synthesis providers compare customer orders against databases of known harmful pathogens and toxins. However, DeepMind warned that AI could generate novel DNA sequences capable of performing similar biological functions without matching existing databases closely enough to trigger current screening systems.

To address this challenge, researchers are exploring ways to adapt DeepMind’s SynthID watermarking technology for biological sequences. The project remains in the research stage and has not yet entered practical deployment.

The company is also investigating advanced screening methods that could identify potentially harmful DNA based on biological function rather than sequence similarity alone.

Improving Outbreak Detection

DeepMind believes metagenomic sequencing could significantly improve early outbreak detection by identifying all microorganisms within a biological sample instead of searching only for known pathogens.

The company noted that high sequencing costs remain one of the biggest obstacles to deploying this technology on a global scale.

To improve efficiency, Google has collaborated with Pacific Biosciences to use AlphaEvolve, an AI coding agent that enhances sequencing accuracy. Researchers are also exploring whether AlphaGenome can help identify pathogens directly from genetic sequence data.

These technologies remain under development and have not yet been deployed as large-scale public health monitoring systems.

Accelerating Medical Countermeasures

The response component of the initiative focuses on accelerating vaccines, diagnostics and treatments for emerging infectious diseases.

DeepMind said more than 10,000 scientific publications have referenced AlphaFold during the past five years, supporting research into diseases such as tuberculosis, malaria, Mpox and Nipah virus.

The company has also partnered with Lawrence Livermore National Laboratory to use AlphaFold 3 for broad-spectrum antibody research, including efforts to develop antibodies against multiple filoviruses.

DeepMind confirmed that it will continue expanding the AlphaFold Protein Structure Database by adding new protein structures relevant to vaccine and therapeutic development.

Meanwhile, Isomorphic Labs has established a dedicated outbreak response unit capable of rapidly deploying its AI-powered drug discovery platform during future disease emergencies.

The company also pledged $7 million to the Health for Human Potential initiative under the Philanthropy Asia Alliance to support infectious disease research across Asia.

Policy Recommendations

DeepMind also outlined several policy recommendations for US lawmakers to strengthen national biosecurity.

The proposals include establishing a federal AI safety framework, introducing mandatory DNA synthesis screening, expanding metagenomic sequencing infrastructure, improving biological data-sharing systems and increasing investment in rapid vaccine manufacturing and emergency clinical trial networks.

Most of these proposals still require legislative approval, meaning many elements of the broader bioresilience strategy remain in the planning stage.

As AI capabilities continue to advance, Google DeepMind and Isomorphic Labs argue that close collaboration between governments, researchers and technology companies will play a critical role in ensuring artificial intelligence strengthens global health security without creating new biological risks.