Evo2 AI model raises bioethics concerns over DNA editing capabilities
Evo2, a generative AI model developed by Arc Institute with Nvidia, Stanford University, and UC Berkeley and San Francisco, can analyze, rewrite DNA sequences and design new genomes. The system trained on 9 trillion DNA nucleotides from 128,000 genomes of diverse species reportedly achieves over 90% accuracy in identifying harmful genetic mutations, but raises bioethics questions about potential misuse in creating artificial life forms.
What happened
Evo2, a generative AI model developed by Arc Institute with Nvidia, Stanford University, and UC Berkeley and San Francisco, was described in a Nature study in March 2026. The system was trained on approximately 9 trillion DNA nucleotides from 128,000 genomes of diverse species, including unicellular and multicellular organisms, plants, and humans.
Evo2 can analyze and rewrite genetic sequences, identify disease-causing mutations, and design new genomes up to one million nucleotides in length. In tests on BRCA1 gene variants, the model achieved over 90% accuracy in distinguishing benign mutations from potentially dangerous ones.
Reported impact
According to developers, Evo2 could reduce time and costs in biological research by accelerating identification of genetic disease causes and development of targeted gene therapies with fewer side effects. The system could potentially replace many preliminary experiments on cells or animals.
Index analysis
Arc Institute co-founder Patrick Hsu stated that Evo2 has 'a generalist understanding of the tree of life that can be used for many applications, from predicting mutations that cause disease to designing possible codes for artificial life forms.' The model represents a significant advancement from its predecessor Evo1, which was trained only on unicellular organism genomes.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.