The Friday Ethics Desk
An AI agent that never denies a prior authorization: where does the accountability sit?
ClarityCare raised $16.5 million for an agent that approves or escalates insurance requests and leaves denials to people. Doctors' groups worry about insurers using AI to deny more care, and a vendor argues the opposite.
By Matteo Favilli, 2026-10-30
In short
Before many treatments, a health plan has to approve the request, and that wait can delay care. A startup called ClarityCare has raised money for an AI agent that speeds the work up and, by design, never issues a denial. Doctors surveyed by the American Medical Association fear that insurers will use AI to deny more care, while an AI vendor says the technology lowers denials.123
What happened
ClarityCare, a healthcare AI startup based in San Francisco, announced a $16.5 million Series A round led by the venture firm Headline. Its funding now totals $19.8 million. The company says it will expand its sales and engineering team, develop new products and grow in the government sector.1
The product is an AI agent called Electra, built for prior authorization: the step in which a health plan decides whether a treatment prescribed by a doctor meets its coverage criteria before the treatment is delivered. According to MedCity News, the system can approve and escalate requests but will never issue denials, and it includes a human in the loop to avoid erroneous decisions. For one of its seven current customers, the company says it cut prior authorization waits from 12 days to two.1
ClarityCare's CEO, Hermine Tranie, describes the process as three parts: assembling the case, the clinical review, and closing the case after the decision. Electra automates the administrative work, strengthens the clinician's review with summaries, recommendations and drafted communications, and completes the paperwork once the clinician has decided.1
The timing follows a federal rule, CMS-0057-F, which sets a decision timeframe of 72 hours for expedited requests and seven days for standard ones. Payers have until January 1, 2027 to meet the requirements.1
The American Medical Association (AMA) has a different view of AI in this area. In a survey of 1,000 physicians reported by the AMA in March 2025, 61% said they fear that payers' use of unregulated AI is increasing prior authorization denials. AMA President Bruce A. Scott said that emerging evidence shows insurers use automated decision-making systems to create systematic batch denials with little or no human review. The AMA also cites a 2024 Senate committee report with AI tools accused of denial rates in some cases 16 times higher than is typical.2
The other side
The case for AI in prior authorization starts with the cost of the current process. In the same AMA survey, 82% of physicians said prior authorization sometimes leads to patients abandoning treatment, and 29% said it led to a serious adverse event for a patient in their care. Physicians and their staff spend on average 13 hours a week on the prior authorization workload for a single physician.2 MedCity News adds an AMA estimate that clinic staff can spend up to two business days per week on gathering records, checking eligibility and filing paperwork, and notes that the AMA itself calls for the promotion of prior authorization automation.1
Brian Covino, chief medical officer of Cohere Health, another company in this field, made the argument directly in an interview with Medical Economics in August. He said physicians' fear is understandable, and that used correctly AI should lower denial rates. His statement was: "Artificial intelligence is never used to deny care." He said a significant portion of denials come from missing information, that AI reading a submission can ask for the missing item in real time, and that Cohere has seen decreases of 30% to 40% in missing-information issues from these prompts. He also said that the company's audits show its AI to be more accurate than humans once fully developed and trained.3 These are claims by a vendor, and the interview gives no independent data.
Matteo's take
Both sides have a point here, and they are describing different tools.
The AMA's worry is about systems that decide against the patient, at scale, with little or no human review. The vendors in these two articles describe systems that sit on the other side of that line: Electra never denies, and Cohere's chief medical officer says AI is never used to deny care. If that holds, the batch-denial scenario in the AMA's survey does not apply to these products. The administrative burden the AMA documents, 13 hours a week per physician, is real and a faster process helps the patient who is waiting.
The open points are about what sits behind the word "approve." An agent that approves the clear cases and escalates the rest also decides which requests a person looks at, and a clinician who receives a summary and a recommendation may follow it. The evidence offered so far is a drop from 12 days to two at one of seven customers, reported by the company, and audit results reported by a vendor. I could not find independent measurement of denial rates in either article.
From these sources, I would say this use is fair when a few conditions hold. Patients and doctors know that AI prepared the case. A named clinician or the health plan stays accountable for each decision, and the system's recommendations are logged so that they can be reviewed. And a patient who receives a denial has a way to contest it and reach a person. The second is part of ClarityCare's design, since a clinician makes the decision. The first and the third depend on the health plan that deploys it.
Here is the question I am left with. If an AI agent approves requests quickly and a human keeps every denial, how would a patient or a regulator ever find out whether the agent was quietly sending too many borderline cases toward a denial?
Sources
Drafted with AI from the sources below.
- ClarityCare Raises $16.5M to Speed Up Prior Authorizations, MedCity News
- How AI is leading to more prior authorization denials, American Medical Association
- 'AI is never used to deny care': on tech, trust, accuracy and prior auths, Medical Economics
All of AI Bits