AI trained on 9 trillion nucleotides designed 16 previously unseen viruses, experts warn

A generative artificial intelligence trained on a dataset of roughly 9 trillion nucleotides produced 16 viral sequences that do not match any known natural viruses, according to reporting. The development has prompted urgent warnings from experts about the pace of biological generative models outstripping current safety measures.

Researchers built the model to learn statistical patterns in DNA. Once it internalized those patterns it was able to output novel sequences that, while synthetic, resemble virus genomes in structure. The sequences were reported as never-before-seen in nature, raising technical and ethical questions about how such tools should be developed and governed.

Why experts are sounding the alarm

Specialists emphasize that the capability to write plausible viral sequences creates dual‑use risks: the same methods that accelerate basic research could be repurposed to design harmful pathogens if left unchecked. Existing laboratory rules and software safety protocols are not universally suited to address generative models that operate at scale.

Priority guardrails for biological AI

• Dual-use risk: models that generate plausible viral genomes could be misused for harmful research.
• Regulation gap: current biosafety and software controls lag behind generative biological AI capabilities.
• Access limits: stricter controls on model training data and output access are needed.
• Audit and red-teaming: independent testing and adversarial review should be standard before release.
• Transparency: funding, methods and oversight arrangements must be disclosed to assess risk.
• International cooperation: cross-border norms and emergency reporting are essential.
• Safer-by-design: developers should prioritize safety features and publication policies that limit risky outputs.

What policymakers and researchers should do next

Policymakers, funders and platform operators should accelerate coordination on access controls, auditing standards and emergency reporting mechanisms. Researchers should adopt safer-by-design practices and clear publication policies to balance scientific value against risk. The story underscores the need for swift, proportionate guardrails around powerful biological AI.

Balancing transparency and security will be difficult: open scientific methods speed discovery but may also reveal how to reproduce risky outputs. Deliberate policies are required to decide what research is appropriate for open publication, what should be shared under controlled agreements, and how to screen computational outputs for hazardous content before dissemination.

The pace of biological generative AI is forcing a hard question: can society capture the scientific upside without enabling serious harm?

As AI capabilities expand into biology, Nuvostella continues to track the governance, safety and responsible-deployment questions that matter for organisations building with advanced AI systems.

Resources & links

• Tom’s Hardware — AI creates 16 new viruses after learning DNA patterns from 9 trillion nucleotides
• CNN — AI creates 16 new viruses from scratch, drawing misuse warnings
• Scientific American — AI created a virus not found in nature, and scientists are worried
• bioRxiv — Generative design of novel bacteriophages with genome language models