Only 7% of candidates selected by AI matched those chosen by human program directors — a statistic cited by Bachina, Hoffmann, and Goodman in a JAMA viewpoint examining the rapid rollout of AI in residency screening.

Writing in JAMA, Bachina, Hoffmann, and Goodman lay out a concerning picture of how fast AI moved into residency application review and how little legal or ethical scaffolding followed. The AAMC's partnership with Thalamus put LLM-based transcript processing, personal statement analysis, and "Academic Career Interest" badges directly into the application platform — free, frictionless, and already live this cycle. The authors argue these risks are not purely hypothetical, pointing to current use and documented transcript-processing errors. Programs may be relying on outputs that Thalamus says are less accurate or may not be displayed when confidence is low, particularly for foreign or low-resolution transcripts.

Here's what's underappreciated: the authors suggest that this isn't just a fairness problem, but a compounding liability problem and a systems-level risk. When LLM-generated outputs (say, a badge or a percentile rank) become inputs to a program's internal scoring algorithm, any upstream bias doesn't just affect one applicant, it propagates systematically across every participating program. The authors flag that a 2025 9th Circuit ruling — Spatz v. Regents of UC — treated residency selection as an employment practice, potentially bringing it under disparate-impact discrimination law. This may raise particular concerns for older applicants and international graduates given known accuracy gaps.

The friction is real: the authors note that even successful legal challenges would come too late to reverse a flawed match outcome. The legal framework remains unsettled, and the liability chain is murky (Thalamus frames its tools as "informational aids," not decision-makers).

They recommend that programs consult legal counsel before deploying these tools and push for stronger oversight from the AAMC. Without clearer standards, they warn that efficiency gains may come at the cost of fairness and increased litigation risk.

Disclosures: Bachina is a fourth-year medical student currently applying to residency in the 2025–2026 cycle. No other disclosures reported.

Source: JAMA