A team research project through the Algoverse AI Research Program investigating whether genomic foundation models are actually robust, or just confidently wrong. The interest traces back to my Biosecurity Atlas, an earlier project mapping the biosafety-and-AI landscape, which got me asking how much we should actually trust these models.

• Built a model-agnostic robustness pipeline (replicate, attack, biological-validity gates, interpret, harden, transfer) spanning seven classification tasks and six DNA/RNA foundation models (AIDO.RNA-1.6B, Nucleotide Transformer, mRNA-FM, CodonBERT, CodonFM, structRFM)
• Designed constrained black-box adversarial attacks (synonymous, structure-preserving) gated by an independent biological-validity oracle, benchmarked against matched random-edit floors
• Built a mechanistic interpretability suite (in-silico mutagenesis, activation patching, concept probes, representation geometry); causal patching revealed sparse-but-redundant encoding
• Delivered LoRA-based defenses that roughly halved attack success, and wrapped the full workflow into a modular agentic research skill

Submitted to a workshop in August 2026; not yet peer-reviewed.