AI Scientist Designs and Runs DNA Experiments
Shanghai Jiao Tong University researchers describe AutoDNA, a system that plans and executes laboratory work with minimal human direction.
Updated on Nov. 4, 2026 in Artificial Intelligence
Live Poll
Do you think AI systems running scientific experiments will benefit society?
Researchers at Shanghai Jiao Tong University announced AutoDNA, an AI scientist system that designs and executes experiments according to researchers' instructions, published in Nature on November 5. The system connects over 20 types of equipment to carry out work from DNA synthesis through genetic variation analysis.
Why it matters
The system's ability to handle experimental execution points to how AI may take on more of the research process. South Korea has set a 2035 goal for an AI that autonomously performs the full research process, underscoring growing government interest in the approach.
AutoDNA is built from multiple AI agents based on Google Gemini 2.5 Pro and adjusted reaction times, reagent concentrations, and washing methods to optimize synthesis. In one run, 14 equipment errors occurred, with the AI resolving 8 autonomously and humans resolving 6.
The players
Shanghai Jiao Tong University
A major research university in Shanghai, China, whose research team developed the AutoDNA system.
South Korea's Ministry of Science and ICT
The South Korean government ministry that set a 2035 goal for an AI scientist under the K-Moonshot program.
The details
AutoDNA uses several AI agents with distinct roles to plan and run experiments. In a test encoding the English sentence 'Intelligence is the ability to adapt to change' into 78 DNA strands, the system completed the work over 203 hours.
Timeline
May: South Korea appointed the inaugural head of its AI scientist mission.
End of September: The Ministry of Science and ICT appointed a new AI scientist mission head.
November 5: Shanghai Jiao Tong University announced AutoDNA in Nature.
2035: South Korea's target for an autonomous AI scientist under K-Moonshot.
The Tech Race
AutoDNA demonstrates autonomous experimental capability in a single laboratory system, which South Korea's K-Moonshot program targets for 2035. The result places the research effort in a broader race among governments and universities to build AI systems that can run science itself.
For researchers, systems like AutoDNA could shorten the time needed for routine laboratory work such as synthesis and sequencing. Scientists using such tools would still need to supervise equipment errors, which in this test required human intervention for six of fourteen failures.
The takeaway
AI systems that plan and run laboratory work are moving from concept to demonstrated practice, but human oversight remains necessary for equipment failures. Readers following the field should watch whether results like AutoDNA's hold up across other types of experiments.
Further reading
Explore more on Artificial Intelligence to see how autonomous research systems are developing across the field.
Live Poll
Do you think AI systems running scientific experiments will benefit society?
