Experts say the lack of detailed methods and source code in Google's AI research for predicting breast cancer “undermines scientific value”
Back in January, Google Health, the branch of Google focused on health-related research, clinical tools, and partnerships for health care services …
Context & Ripple Effects
Google's healthcare ambitions have always run through credibility: the Medical Brain team's disease-prediction work was the original case for why Google belonged in medicine at all. The breast cancer model was the flagship of that pitch from Google Health — which makes the experts' charge that missing methods and source code undermine its scientific value a direct hit on the franchise's founding claim.
It also extends a pattern the related coverage documents: frontline clinicians testing Google's AI nurse handoff tool raised consistency and quality questions in 2024, and in early 2026 Google had to pull AI Overviews for liver health queries after experts flagged alarming outputs. Repeatedly, external scrutiny is doing the validation work Google's own releases don't.
First-order effects
- Clinicians and researchers cannot independently verify or build on the breast cancer model without methods and source code, capping its path from paper to clinical tool and weakening Google Health's evidence base for hospital partnerships.
Second-order effects
- The reproducibility critique hands an opening to rivals willing to publish openly — pressure that aligns with Google's own later move to release open TxGemma models for drug discovery, suggesting the criticism registered internally.
Third-order effects
- If closed-model health AI keeps failing expert review while open releases gain traction, publication norms and purchaser requirements in medical AI will shift toward mandatory code and methods disclosure as a condition of adoption.
The trend: Health AI is moving from headline-grabbing closed results toward openness and external verification as the currency of clinical trust.