A top ten U.S. hospital is letting a vendor's AI handle 24% of its patient bill collections. The hospital won't name itself, the metrics aren't audited, and the patient isn't told the voice is software.
A patient, three days out of a hospital stay, picks up a ringing phone. The voice on the other end introduces itself as a representative of the hospital's billing department and walks through a payment plan. The voice does not say it is software, because the vendor that built it has decided the patient does not need to know that, and the hospital that licensed the technology has not required the vendor to say so. That is the new front line of American medical debt, and the people building it are not required to tell you where the line is.
Intelligent Contacts said on June 15 that its Grace AI agent has moved to general release after roughly a year of live beta, and that a top-ten U.S. healthcare system (the company will not name it) is now running 24% of all patient payments through Grace with no human collector in the loop, two months after going live. The press release is also the source of every number attached to that claim, including 63% more payments collected, 47% more accounts worked, and 60% faster collection cycles, all measured by the vendor against the same vendor's earlier deployment, with no independent audit cited.
The "top-ten" framing is also the vendor's. Intelligent Contacts did not say whether it meant U.S. News & World Report's Honor Roll, bed count, operating revenue, or some other ranking. A reader who wants to verify the result cannot call the hospital and ask, because the hospital is anonymous, and a reader who wants to check the methodology cannot, because the methodology is not published. What the press release calls evidence is, in form, a product announcement.
Grace is being sold as the first AI collection agent to fully replace a live agent across every stage of collections: opening the call, negotiating a payment plan, handling a dispute, closing a discounted settlement. That is a marketing position, not a benchmark. Other vendors in healthcare-focused voice AI, including Syllable, Hippocratic AI, and several contact-center platforms that have moved into clinical revenue cycle work, all describe comparable end-to-end handling for at least some stages of collections, and none of those claims has been independently verified either. The "first" framing is unfalsifiable on the evidence in the release.
The compliance language in the announcement is where the gap between federal law and certification is widest. Intelligent Contacts writes that Grace operates "in complete compliance with FDCPA, FCRA, TCPA, and HIPAA." That sentence lists four federal statutes any debt collector must obey: the Fair Debt Collection Practices Act, which restricts how and when collectors may contact consumers; the Fair Credit Reporting Act, which governs what collectors may report to credit bureaus; the Telephone Consumer Protection Act, which regulates automated dialing and prerecorded calls; and the Health Insurance Portability and Accountability Act, which protects patient health information. Listing them is not the same as being audited against them, and no outside counsel, regulator, or accreditation body is cited as having reviewed Grace's scripts, escalation logic, or call recordings.
The patient-side question is the one the press release does not address. Under the FDCPA's "mini-Miranda" rule, a debt collector must disclose, at the start of a call, that the call is an attempt to collect a debt and that anything the consumer says may be used for that purpose. Whether an AI-initiated call satisfies that rule, who on the call is responsible for delivering the disclosure, and whether the patient is told they are speaking to a machine, are not settled in the release. Those are questions for the hospital's compliance officer, the vendor's general counsel, and, eventually, a regulator or a court. None of those answers is in the record.
What hospital finance and compliance officers can do this week is treat the press release as a starting brief, not a procurement document. Ask any vendor pitching an AI collections agent for three things in writing: independent efficacy data with a defined baseline, a patient-side disclosure script that names the AI and delivers the FDCPA notice on every call, and a complaint-escalation path that puts a human in the loop within a stated number of minutes. The "first" claim and the "complete compliance" claim are both things a vendor will say; they are not the same as proof.