Blue Cross: Hospital AI Added $942 Million in Bills — Sep 25, 2026

0:00Hospital AI added nearly a billion dollars in charges. Nobody got sicker. Welcome to MedAI Times podcast, your daily update on medical AI. Don't forget to like and subscribe. Blue Cross insurers say Hospital AI added nearly a billion dollars in charges with no sicker patients to show for it.

0:18The Veterans Health Administration picks a bridge for ambient AI nationwide on a contract worth up to $776 million. Minnesota doctors end the first inpatient physician strike at a private American hospital with AI on the bargaining table.

0:35Anthropic and open evidence give away clinical AI in about 100 countries. Claude claims a new enzyme system and CRISPR researchers shrug. A smartphone app beats the standard test for telling Crohn's disease from intestinal tuberculosis and three Frontier labs cut their prices in the same week.

0:54Last Friday, we covered Medicare's AI denying prior authorizations. This week, the insurers say AI on the other side of the desk is inflating the bills. Start with the money. The Blue Cross Blue Shield Association analyzed its own members inpatient claims this week.

1:10Between 2024 and 2025, rising coding intensity added about $942 million in cost compared with 2023. 653 million of that traced to secondary conditions, the extra diagnoses that sit alongside the main reason a patient was admitted.

1:29Over the same stretch, the share of Blue members inpatient claims billed as complex rose from roughly 37% to about 40%. What did not change, the association says, is the care.

1:42Their medical executive put it flatly. AI is finding more billable conditions, not sicker patients. The tools named are ambient listening scribes and lab data mining. Here's the mechanism. One documented comorbidity can push a claim into a higher-weighted payment tier without a single order changing.

2:02The scribe is not inventing anything. It captures what was genuinely said in the room and what was said is not the same as what was treated. Signal with an asterisk. This is an insurer's own analysis, not a peer-reviewed study, and insurers have an obvious interest in lower payouts.

2:20But it covers three years of real bills. Which brings us to that purchase. On September 22nd, a bridge announced it was selected under a new Veterans Affairs Enterprise contract with a total ceiling of $775.7 million over five years across all eligible vendors.

2:39The tool runs across both VA record systems, the old VISTA and the newer federal one, covering more than nine million veterans. A bridge was already live at more than 75 VA medical centers through a competitive pilot.

2:53One thing worth saying out loud, the VA pays salaries, not fee-for-service claims. The billing incentive that drives the Blue Cross finding is much weaker there. Same technology, different economics.

3:06In Minnesota, about 150 doctors at Alena Health's Mercy Hospital in Coon Rapids and its Unity Campus in Fridley reached a tentative first contract this week after a four-day strike that ended the week before.

3:20It is believed to be the first strike by inpatient physicians at a private American hospital. Three years of bargaining, more than 66 sessions. The three-year deal includes pay transparency, greater professional autonomy, and a grievance process for advocating on behalf of patients.

3:38The union dropped its paid sick leave demand. AI was one of the central fights. The doctors objected to software suggesting diagnostic codes and billing for their patients. The contract still has to be ratified.

3:51Anthropic and Open Evidence said this week they will offer a free, locally-adapted version of Open Evidence to physicians in roughly 100 low- and middle-income countries, including Uganda, Angola, Sudan, Haiti, and Mongolia.

4:05Anthropic supplies the models. Open Evidence tunes the tool for regional disease patterns and for what diagnostics are actually available. Anthropic president Daniella Amodei said market incentives alone would not make this happen.

4:20Fair. What we do not have yet is outcome data from any of those settings, and the closest thing to it, the PENDA Health trial in Kenya, was a different tool entirely. Anthropic also says Claude found something new in biology.

4:34950 agents ran for about 21 hours, burned 210 million tokens, pulled together more than 200,000 reverse transcriptase enzymes, and flagged 3,500 candidate systems.

4:47The pick of the litter is a structure Anthropic calls array-associated reverse transcriptases, found mostly in the viruses that infect bacteria. Sitting beside a run of evenly-spaced DNA repeats that looks a bit like a CRISPR array.

5:03Feng Zhang reviewed it and called it intriguing and worth investigating. Lucas Harrington, who trained in Jennifer Doudna's lab, called it routine genome mining and noted similar systems have been known since 2008.

5:17Anthropic is candid that it does not yet know what the thing does, which is a slightly awkward place to plant a flag. One proper peer-reviewed result this week. In elementary pharmacology and therapeutics on September 21st, a team led from the All India Institute of Medical Sciences in New Delhi built a phone app to separate Crohn's disease from intestinal tuberculosis,

5:41which look nearly identical on scope and get treated in opposite directions. Trained on 1,066 patients, then validated prospectively on 121. The model scored 0.921 on the standard accuracy curve against 0.750 for the Bayesian reference clinicians use now.

6:00One center, small prospective cohort, but it is a real prospective test of a real bedside problem. On the general AI side, the same 48 hours saw Anthropic ship Claude Opus 5.5 and OpenAI ship GPT-6 Sol and Luna.

6:17Opus 5.5 matches Fable 5.1 on most work at 40% lower cost with input at $4 per million tokens. OpenAI roughly halved its prices too. The medically interesting part is Anthropic's, Opus 5.5's advanced biology capability is gated behind a new life sciences verification program.

6:36So you have to be a vetted lab or pharma company to get at it. Spotlight coding intensity. Every inpatient stay gets mapped to a payment group and adding one documented comorbidity can bump that stay into a higher paying tier without changing a single order.

6:53Ambient scribes and lab mining tools are extremely good at surfacing conditions a busy clinician would not have typed, which is exactly what makes them valuable and exactly what makes this week's billing number hard to interpret.

7:07Every study, contract number and report behind this episode is linked below, including the Blue Cross analysis so you can read their methodology yourself. Has your ambient scribe ever put a diagnosis in your note that you never actually treated?

7:22Thanks for listening. Find us on YouTube and your favorite podcast app. See you tomorrow.