From prompts to genomes: A constraint-first framework for AI-driven sequence design across life and viruses

A constraint-first framework for future 'prompt-to-genome' AI across Bacteria, Archaea, Eukaryota, and viruses is proposed. Genome-scale models can now generate long DNA sequences, and experimentally viable genome-scale designs have begun to appear, but sequence generation alone does not establish biological feasibility. Constraints should not require designs to remain within natural sequence space; rather, they should define testable boundaries for function, inheritance, lifecycle coherence and safe use, including new-to-nature designs. We propose a two-part system. First, a Constraint Compiler translates natural-language prompts into machine-checkable specifications defining scope, context, objectives, assumptions and governance requirements. Second, a modular Validator Stack (C1-C6) evaluates candidate designs for sequence integrity, gene and regulatory plausibility, genome completeness, comparative coherence, systems-level feasibility, and safety and governance. Extreme prompts serve as stress tests to expose hidden assumptions, conflicting objectives, and refusal points, complemented by adversarial testing and risk assessment. A phased roadmap and community benchmarks are proposed to measure specification accuracy, biological feasibility, robustness, and governance performance. The goal is to move prompt-to-genome systems from impressive generation toward reliable, auditable, and accountable biological design.