AutoDNA: The AI Lab That Runs Its Own Experiments

Bold gradient image in crimson, amber and indigo showing an AI laboratory with robotic arms running DNA experiments.

 

Imagine describing an experiment in plain English, walking away, and coming back to find it done — the robots planned it, ran it, and fixed their own mistakes. That is AutoDNA, an AI-native laboratory built at Shanghai Jiao Tong University, published as a peer-reviewed paper in Nature on 4 November 2026.

The project was led by computer scientist Haibo Chen, with Mingyu Wu, Zhaoguo Wang and Jiabin Wang. A preprint appeared on arXiv in July 2025, but the Nature paper is the full, reviewed account of what the system can do.

What is AutoDNA?

AutoDNA is a laboratory run by AI agents. Instead of a scientist pipetting samples by hand, software agents take a request written in plain language, plan the experiment, and drive robotic arms across more than 20 types of laboratory instrument.

A single experiment can run for more than 11,000 hardware steps, covering synthesis, transcription, amplification and sequencing. Humans are still involved, but only for three jobs: refilling reagents, replacing labware and servicing the hardware. Everything else is handled by the system.

What did it actually prove?

The Nature paper describes four demonstrations, and each one tests a different part of the system.

First, the team ran an end-to-end nucleic-acid testing service — the kind of test used to detect genetic material in a sample — from start to finish without human intervention in the experiment itself.

Second, the system optimised enzymatic DNA synthesis on its own, reaching yields comparable to expert-optimised work, without an expert in the loop.

Third, it ran several protocols at the same time, finishing 3.2 times faster than running them one after another.

Fourth, the system used DNA itself to store data: it encoded a sentence into 78 DNA strands and read it back without error, in a run of 11,382 hardware steps over about 203.5 hours (roughly eight and a half days).

There was a fifth result tucked into the paper as well: the system correctly identified a KRAS G13D mutation in a real clinical sample. KRAS is a gene linked to several cancers, and spotting its mutations matters for diagnosis and treatment decisions.

It makes mistakes — and fixes some of them

Fourteen hardware errors were recorded during the experiments. The system fixed eight of them on its own and needed a person for the other six. A lab that can recover from more than half of its own errors is genuinely useful, but it is not yet a lab you can leave alone.

What are the limits?

AutoDNA's limits are just as important as its achievements, and the team is open about them.

People must still replenish reagents and swap labware, so the lab cannot run indefinitely without attention.

The DNA-storage run was a controlled demonstration, not proof that DNA is ready for large-scale archival use. Storing one sentence over eight days is a long way from storing a library, and the team does not claim otherwise.

The KRAS detection validates the system, but it is not a new clinical assay. Finding a known mutation in one real sample shows the pipeline works; it does not mean AutoDNA is ready to be used as a diagnostic tool in hospitals.

The team says that using AutoDNA on living systems will need a dedicated biosafety and biosecurity plan, with guardrails, human-authorisation steps and the ability to monitor and pause experiments.

Why it matters

AutoDNA points to a different kind of laboratory. Today, a skilled scientist spends much of their time on the physical work of experiments. If AI agents can take over that work reliably, scientists can spend their time on the questions that matter: what to test, and what the results mean.

Key takeaways

  • AutoDNA is an AI-native lab at Shanghai Jiao Tong University, published in Nature on 4 Nov 2026.
  • AI agents plan experiments in plain language and drive robots across 20+ instrument types and 11,000+ steps.
  • It optimised DNA synthesis to expert-level yields and ran concurrent protocols 3.2x faster.
  • It stored a sentence in 78 DNA strands and read it back error-free in about 203.5 hours.
  • It correctly identified a KRAS G13D mutation in a real clinical sample.
  • Of 14 hardware errors, the system fixed 8 itself and needed a person for 6.
  • Limits: humans must refill reagents and swap labware, and living-system work will need a biosafety plan.

Conclusion

AutoDNA is not a replacement for scientists. It is a laboratory that does the physical work of science so scientists can focus on the thinking.

FAQ

What is AutoDNA?

AutoDNA is an AI-native laboratory from Shanghai Jiao Tong University. AI agents take a plain-language request, plan the experiment, and drive robotic arms across more than 20 instrument types. Humans only refill reagents, replace labware and service hardware.

What did AutoDNA prove in its Nature paper?

It ran an end-to-end nucleic-acid test, optimised DNA synthesis to expert-level yields, ran concurrent protocols 3.2x faster, stored a sentence in 78 DNA strands and read it back without error, and identified a KRAS G13D mutation in a real sample.

Can AutoDNA run completely on its own?

Not yet. People must still replenish reagents and swap labware, so it cannot run indefinitely unattended. It also fixed only 8 of 14 hardware errors by itself, and work on living systems will need a biosafety and biosecurity plan.

Sources

Related reading on Daily Tech & AI:

Researched with AI assistance and reviewed for accuracy.

Follow on WhatsApp

Follow Daily Tech & AI on WhatsApp for the latest tech and AI updates: https://www.whatsapp.com/channel/0029Vb9XzdHDJ6H4TFNAl61r

Comments