The Frontier Model Forum (FMF) has a founding mandate to advance the science of frontier AI safety and security. As part of that effort, today we are pleased to share an update on our support for novel research at the intersection of AI and biological sciences. 

Virology and Bacterial Biothreat Benchmarks

Through its work with the AI Safety Fund, the FMF has supported the development of novel virology and bacterial biothreat benchmarks. Completed earlier this year, these include:  

• Virology Capabilities Test (VCT) – A critical question facing AI developers and regulators is whether frontier models could enable malicious actors to conduct dangerous virology work. To address this, SecureBio‘s AIxBio team developed the VCT benchmark—an expert-validated assessment tool for evaluating AI models’ ability to provide practical troubleshooting guidance for viral research.

This benchmark features 322 multimodal questions covering dual-use research methods, complete with original micrographs and photographs. The dataset is now available to safety teams, regulatory bodies, and academic researchers working on AI-bio safety.

• Bacterial Biothreat Benchmark Generation (BBG) – To strengthen biothreat benchmarks for AI model safety assessments, Nemesys Insights developed the Bacterial Biothreat Benchmark Generation (BBG) Framework, a comprehensive assessment system aligned with biosecurity threat chains.
  
  The goal of the framework is to enable the identification of key areas of potential harm along the biothreat chain, allowing for the prioritization of mitigation efforts and enhancing safety more than is possible with more general approaches. The final framework contains approximately 1,000 high-quality bacterial biothreat prompts, complete with implementation tools and risk mitigation recommendations. The benchmark dataset and implementation tool, as well as associated research publications, are forthcoming, pending an information hazard review.

FMF and Sentinel Bio Wet Lab Study

FMF, together with Sentinel Bio, a non-profit committed to pandemic prevention, is funding a large-scale wet lab study to help assess whether frontier AI systems can aid or “uplift” non-experts to carry out expert-level biological tasks. This study is one of the most comprehensive wet lab studies of its kind in terms of duration and number of participants. It is an important step in understanding the intersection of AI capabilities and biological research, essential for mapping misuse risks and building appropriate model-level safeguards.

The study will evaluate whether access to a leading multimodal AI model improves the success rate of novice lab participants in completing a defined, multi-step Biosafety Level 2 (BSL-2) laboratory procedure. Participants will be divided into two groups: one with AI assistance and one control group working with internet-only access. By comparing their performance, the study aims to assess whether AI can meaningfully uplift the wet lab capabilities of novices in advanced biological research. Selected participants will go through an institutional review board consent process. 

Given the sensitivity of any potential findings, all results will undergo pre-review by an independent advisory board of biosafety and national security experts prior to publication. The decision and timeline for release of the results will be dictated by this pre-review process. Over the next few months, we plan to provide more updates on our grant-making work to support the science of AI safety.

For any questions or inquiries, please contact aisf@frontiermodelforum.org.