# Preprint finds biological design tools increase AI mutation-design risk scores

_Published Monday, October 12, 2026 at 3:10 PM EDT · AI, Science · Latest · Tier 2 — Notable_

Researchers report that specialized biological design tools increased the risk scores of AI-generated mutation designs in DRYLAB-Bench, a new benchmark evaluated on 15 frontier models. The preprint tests agents across 30 hazardous mutation-design scenarios and scores their designs against 187,695 existing experimental measurements from ProteinGym and ViroGym.

Without tools, 67.6% of one-shot designs exceeded the average measured variant, and 21.4% reached the highest-risk decile. Adding tools raised the pooled Biosecurity Risk Score by +0.116 and increased the top-decile share to 23.5%. The evaluation uses existing measurements to assess designs rather than experimentally testing newly generated biological modifications.

## Sources

- [cs.AI updates on arXiv.org](https://arxiv.org/abs/2610.13116)

---
Canonical: https://techandbusiness.org/newswire/TPLIg5PWMto9j0XWTClsbQ
Published: 2026-10-12T19:10:20.717Z
Story chronology: 2026-10-12T04:00:00.000Z
Retrieved: 2026-10-12T21:47:13.755Z
Publisher: Tech & Business (techandbusiness.org)