Latest from the FMF: Grant-Making to Address AI-Bio Risk Challenges
The Frontier Model Forum has announced new research initiatives addressing AI and biological science safety. The organization has supported development of evaluation tools through its AI Safety Fund.
Two benchmarking projects were completed:
Virology Capabilities Test (VCT) - SecureBio's AIxBio team created an assessment containing "322 multimodal questions covering dual-use research methods" to evaluate whether frontier AI models could enable dangerous virology work. The benchmark includes original microscopy images and photographs, and is now available to safety teams and researchers.
Bacterial Biothreat Benchmark Generation (BBG) - Nemesys Insights developed a framework to identify potential harms along biothreat chains. The system contains approximately 1,000 high-quality bacterial biothreat prompts with implementation tools and risk recommendations. Results are pending information hazard review before release.
Wet Lab Study - The FMF is partnering with Sentinel Bio to fund a large-scale laboratory study examining whether frontier AI systems can help non-experts perform expert-level biological work. The research will divide participants into two groups—one with AI assistance and one with internet-only access—to evaluate performance on defined Biosafety Level 2 procedures. All findings will undergo pre-review by independent biosafety and security experts before publication. Additional updates on grant-making efforts are planned for the coming months.