A writer who tested an AI’s ability to analyze a full human genome in 30 minutes found that the technology could replicate key findings from a decade-old expert review but also exposed gaps in how these tools handle incomplete or outdated genetic data.
The experiment began with a simple prompt: ask an AI to analyze an entire genome file, including rare variants linked to inherited disorders, drug response genes, and disease risk scores. The task mirrored a 2009 effort that required 30 researchers, a custom-built analysis engine, and nearly a year of work to interpret the genome of a colleague. That study, published in The Lancet, was one of the first to translate a full genome sequence into personalized medical advice.
This time, the same genome—sequenced in 2009 using an older reference and lacking modern data—was uploaded to Claude, an AI system. Within minutes, it flagged the same critical finding: the writer carries two copies of the APOE ε4 variant, a well-known genetic risk factor for Alzheimer’s disease. It also identified variants in the DPYD and CYP2C19 genes, which influence drug metabolism, matching the original team’s conclusions. When asked to estimate disease risks, however, the AI declined, citing insufficient data in the file.
Read Also: U.S. to cut Medicaid for legal immigrants starting October
The entire process used about 400,000 tokens and cost roughly $5. The bulk of the half-hour runtime was spent waiting for scientific databases to load, a far cry from the months of manual review required in 2009.
Yet the experiment also revealed how easily modern tools can mislead. The genome file was generated using short-read sequencing, a common but imperfect method that struggles in regions where DNA sequences repeat or overlap. Disease-causing variants can hide in these areas, leaving some analyses incomplete. Worse, most genome comparisons still rely on a reference sequence built from a limited set of individuals, making it harder to detect variations in underrepresented populations.
Newer reference genomes now incorporate multiple branching paths to better account for human diversity, but adoption remains uneven across labs and clinical systems.
Read Also: SNAP cuts linked to rise in family separations
For now, the biggest risk isn’t the technology itself; it’s the lack of clear rules for what constitutes a reliable genomic interpretation. A medical-grade test should detect not just simple DNA changes but also complex structural variations, some of which are harder to spot than a misspelled word in a book. The scientific community is still debating how to define those standards, especially as consumers gain direct access to their genetic data.
As AI tools become more accessible, the question isn’t just whether they can analyze genomes faster. It’s whether the results will be trustworthy enough to shape real-life decisions, without leaving users to interpret them alone.
Services like 23andMe have primarily focused on a limited set of common genetic variants. A whole genome contains vastly more information. With AI agents, consumers are no longer limited to a fixed interpretation of their results. They can ask follow-up questions and request additional analyses. This ongoing interaction with one’s own genomic data was not possible before.
