ADAPT: a programme for the advanced detection of AI-enabled pathogenic threats

Advances in AI are expanding both the ceiling and the accessibility of biological engineering, creating threats that existing synthetic nucleic acid screening is not equipped to detect. The IARPA-funded Functional Genomic and Computational Assessment of Threats (FunGCAT) programme advanced screening by creating tools specialised for sequence screening and by progressing on the annotation of potential sequences of concern. However, 3 years after the conclusion of FunGCAT, critical gaps remain: (1) the field lacks an operationalisable definition of what makes a sequence a biosecurity concern, and (2) current tools cannot detect threats on the basis of function rather than sequence similarity. To close these gaps, we propose the Advanced Detection of AI-enabled Pathogenic Threats (ADAPT) programme in two phases as a successor to FunGCAT. ADAPT Phase I would develop a multi-attribute, function-based definition of sequences of concern and generate the benchmark datasets. Phase II would develop and validate screening tools capable of detecting known threats, AI-paraphrased functional homologues, and, where possible, AI-designed novel threats. Continuous governance workstreams would translate technical outputs into regulatory guidance and maintain secure infrastructure. ADAPT builds on FunGCAT's legacy and the subsequent work of the synthetic nucleic acid screening community, while adapting to an era in which biological AI models can generate functional sequences bearing little resemblance to any previously characterised sequence.