AI Is Making a Mess of Nurses’ Schedules. They Say It’s a Safety Issue

WIRED reports nurses across HCA Healthcare — the largest U.S. hospital chain — describe Timpani, an AI scheduling system co-developed with Palantir and rolled out since 2023 to roughly 130 of ~190 locations, as routinely ignoring preferences, stacking exhausting consecutive shifts, and under-weighting experienced staff, especially Sundays.

Critical care nurse and National Nurses United leader Amber Retzloff says over half of 50 requested 12-hour shifts were overridden. HCA spokesperson Harlow Sumerford says nursing leaders — not Timpani — make final decisions and denies the tool is designed to cut staffing at patients’ expense. Innovation exec Michael Schlosser has claimed big cuts in manager scheduling hours, less contract-nurse reliance, higher retention, and that over 98% of schedules mix skills and experience. A former HCA data science manager’s July suit alleges data deletion that blocked audits; HCA had not formally responded in court when WIRED published. Rayus Radiology staff separately reported Palantir Foundry scheduling errors tying orders to wrong patients or inventing details.

We’re pro-tools that give clinicians time back. We’re not pro-black-box rosters that burn the people holding the crash cart. If nurses across multiple states say the same thing — red days that used to be sacred now get scheduled, junior-heavy teams, appeals bouncing to Nashville with no conversation — then “leaders make the final call” is a thin shield over an automated default that’s hard to fight.

HCA’s efficiency claims may be real for managers and labor cost. They can still be false for bedside safety. Retzloff’s account of delaying care for the sickest patients to coach novices because she was the only senior on shift is exactly the skill-mix failure mode algorithmic forecast tools create when “coverage” is a headcount, not a competence map. Barker’s Missouri psych unit grievance and claimed quits/injuries deserve scrutiny in arbitration, not a shrug.

Palantir’s posture — we present information; customers own data and decisions — is legally tidy and morally incomplete when the product is marketed as generating stronger initial schedules. Rayus’s alleged wrong-patient and hallucinated-detail failures are a different failure class: extraction without enough human review on messy inputs.

Useful AI in hospitals starts with auditable objectives nurses help set — honor red days, enforce seniority mix, surface confidence — and with logs that can’t be casually deleted. Until then, I’m with the nurses treating this as a safety issue, not a change-management complaint.

Timpani runs on Palantir Foundry, forecasting patient volume and staffing. National Nurses United represents ~10,000 of HCA’s ~100,000 nurses and says information requests about the tech were rejected. Palantir’s health-care push coincides with soaring sales and separate UK NHS controversy.