When AI Gets Genetics Wrong: A Case Study Series
By Sequencing Team, The team of bioinformaticians, Genetic Health Coaches, and writers at Sequencing.
Jul 30, 2026
Case Studies in This Series
When AI Gets Genetics Wrong: Case 1 — Familiar Names, Wrong Context
The ESPN Gene & Dalmatian Hypouricemia
Two examples where AI anchored on the most common everyday meaning of a word — a sports network and a dog breed — and ignored the actual genetic entity being asked about. Both responses were confident, well-structured, and completely wrong.
When AI Gets Genetics Wrong: Case 2 — Variant Identity Confusion in BRCA1
Google AI Returns Information About the Wrong BRCA1 Variant
A real customer inquiry reveals how Google AI mixed up two distinct BRCA1 variants — one common and clinically benign, one rare and pathogenic — and returned a pathogenic classification for the wrong one, citing databases that actually contradict the claim. This case shows the risks of AI-assisted variant interpretation without access to a person's full genomic profile.
Why This Matters
These cases share a common underlying problem: AI systems pattern-match on surface-level features — familiar words, shared identifiers, adjacent database entries — rather than reasoning through the actual scientific question. The results often look correct at a glance, which makes them more dangerous than an obviously wrong answer.
For genomics and clinical contexts, AI-generated content should always be verified against primary sources such as ClinVar, OMIM, NCBI Gene, UniProt, or peer-reviewed literature before being used in any patient-facing or research context.