JRC flags readiness gaps in biological AI models

JRC research based on a dataset of 480 biological AI models finds rapid progress in data-rich fields such as protein structure, function annotation and molecular design, with slower development in RNA, single-cell and clinical domains where data is more limited and less standardised. The EU has a strong scientific base and high-performance computing capacity, but weaker intra-EU collaboration, fragmented repositories and data governance gaps limit readiness.

High-profile systems including AlphaFold and ESM3 are described as domain-mature but still at low-to-mid technology readiness levels, with no surveyed model subject to an integrated readiness assessment. That gap creates a “maturity paradox” in which benchmark performance does not necessarily translate into certified clinical or industrial deployment, while also raising biosecurity concerns around potential misuse in pathogen design or toxin engineering.

Academia participates in the development of 85% of the surveyed models, while industry participates in nearly 40%; only 17% of industry-only developed models release training code. Policy priorities include stronger biological data infrastructure, support for European biological AI foundation models as public goods, more strategic intra-EU collaboration and clearer frameworks for assessing scientific maturity alongside deployment readiness.