Mayo Clinic AI whistleblower sues over MAYA tool errors

A civil suit against Mayo Clinic is the first one I have seen that puts a specific error-rate number on an internally deployed clinical AI, and that is the part worth sitting with. According to Futurism's write-up of reporting by MPR, former Mayo research director and AI compliance lead Traci Tamiko Eto alleges the team behind MAYA, the hospital's AI-integrated digital assistant, knew the tool had an error rate as high as 67 percent and tried to hide it.

The rest of the complaint tracks a familiar whistleblower shape. Eto says she flagged privacy problems with a separate system, the Mayo Clinic Platform, and that her supervisor pushed back on the grounds that fixing the issue would, in her supervisor's alleged words, "jeopardize the pace on ongoing research projects, which in turn would compromise Mayo's competitive advantage." She says the MAYA team deleted unflattering test results and mischaracterized the tool's abilities. After ten separate whistleblower complaints, per the reporting, she was excluded from executive meetings in early 2025, told she was a "poor cultural fit," and offered a choice between resigning and having her personnel file altered in ways that would, per the suit, "render her unemployable."

The reason this matters beyond one hospital is that clinical AI review is still mostly self-attested. Vendors and internal teams run their own evals, publish their own accuracy numbers, and regulators lean on that paperwork. A live allegation that an internal team suppressed a two-thirds error rate is exactly the fact pattern that invites federal review of how these tools get cleared for patient-facing use.

The honest caveat is that this is one side of a filed complaint, not a proven finding, and Mayo's communications director told MPR the organization's "research and clinical innovation are conducted in accordance with applicable laws and regulations," declining further comment on active litigation. What the reporting does not give you is which court the case sits in, what damages Eto is seeking, or any specific documented harm to a named patient. The 67 percent figure is her characterization of what internal testing allegedly showed, not an independently audited number.

Still, the ground is shifting. If you run compliance at a health system deploying an AI assistant, the useful move this week is not a press statement about your ethics posture, it is checking whether your own internal error logs would survive discovery.