A pricing tool that promised efficiency is helping hospitals bill higher severity codes for the same care, and a Blue Cross audit shows most of those new codes don't hold up clinically.
At the hospitals in a Blue Cross Blue Shield analysis, the billing code for acute posthemorrhagic anemia in new mothers rose from 4% to 12.3% of maternity admissions between 2022 and 2025. The blood transfusion rate, which is what actually happened in those rooms, barely moved. An internal audit of the hospital system with the steepest jump found that fewer than 20% of the newly coded cases met clinical criteria for the diagnosis, according to STAT News reporting cited by Fortune.
That mismatch is the story inside the bigger one. A new wave of ambient AI note-taking tools, the kind that quietly listen to a patient visit and draft the clinical note, is giving providers a more granular view of every encounter. More diagnoses captured, more comorbidities logged, more billable specifics on the page. When that extra detail is fed into the standardized billing labels that tell insurers what to pay, the same visit can land in a higher-paying severity tier without any change in the care delivered.
PwC's 60-page Behind the Numbers 2027 report, released June 11, projects U.S. commercial health costs rising about 9% in 2027, tying for the highest medical cost trend since 2010 and 2011. Nearly 70% of the 27 health plan actuaries PwC surveyed ranked provider AI documentation and coding products as a top-three cost inflator for next year, and roughly 20% called it the number-one inflator, according to Healthcare Dive's write-up of the report.
The pattern is visible in the coding itself, not just in projections. Higher-severity codes pay more. When a system that previously under-documented starts capturing every qualifier an attending physician mutters, the average claim migrates upward. As one health insurer executive told Axios, providers "will take AI and say, 'How can I use this to further my self-interest?'" (Fortune).
PwC's U.S. health industries leader Glenn Hunzinger pushed back on treating AI as the main villain. AI documentation and coding "is not the biggest piece" of the 9% trend, he said. Labor and supply inflation, provider consolidation and contracting pressure (PwC says about 65% of plans flagged it), the No Surprises Act independent dispute resolution process, GLP-1 pharmacy spending, and rising behavioral health demand are the larger drivers. AI is the new pressure, but it is landing on top of an already-tilted table.
The counter-argument is real, just slower. Morgan Stanley projects that AI could eventually deliver trillions of dollars in healthcare savings by 2050 through administrative automation and earlier diagnosis. That is a long-horizon forecast attached to a present-tense audit, and it doesn't cancel the current bill.
The friction already exists. Blue Cross Blue Shield plans and other payers can audit AI-generated notes the same way they audit clinician notes, checking whether a higher-severity code is matched by a corresponding change in treatment, lab values, or patient mix. Regulators can ask the same question. And any patient who sees an unexpected jump in what a visit cost can ask the simpler version out loud: "What severity code did you use this visit, and what in my record justified it?" The system is optimizing for revenue capture faster than for oversight, and the oversight tools, payer audits, patient questions, and clinical review, are already in the room. The question is whether they get used at the same speed as the software.