Blue Cross Blue Shield Association’s Chief Product Officer said the quiet part out loud last week: AI-assisted hospital coding has added an estimated $942 million to the insurer’s costs since 2023, and the pattern doesn’t look like patients getting sicker. It looks like software getting better at finding billable diagnoses.
The $942 Million Number
According to a BCBSA analysis reported on September 24-25, 2026, roughly 70% of that figure — about $650 million — traces to secondary diagnoses added to claims after AI documentation and coding tools flagged them. BCBSA’s Luke Chalker, SVP and Chief Product Officer, framed the concern directly: “The disconnect between diagnoses and treatment suggests that AI is identifying more billable conditions, not sicker patients.” The association estimates roughly 60% of hospitals are now using AI-powered diagnostic or coding-support tools in some form.
Why Secondary Diagnoses Are the Flashpoint
Secondary diagnoses drive DRG assignment and severity-of-illness weighting under Medicare’s inpatient payment system — add a qualifying comorbidity and a claim can jump a full DRG tier without a single additional procedure being performed. That’s exactly the mechanism BCBSA is pointing at: AI tools trained to catch every codeable condition in a chart will, by design, surface more of them than a human coder working under time pressure ever would. The problem isn’t that the codes are wrong on their face — it’s proving, after the fact, that each one is clinically supported and not just statistically plausible.
Hospitals Are Running the Same Playbook
Hospitals aren’t waiting to see how this plays out. Dave Mazurkiewicz, CFO of Michigan-based McLaren Health Care, told reporters his system uses AI specifically because payers do: “All the insurers are using A.I. to scan our charts to look for claims to deny. For the same reason, we’re looking at the same charts today.” Abridge founder and CEO Shiv Rao, whose ambient documentation tools sit on the provider side of this dynamic, put it more bluntly: “It’s bots fighting bots, agents fighting agents, a horrible dystopic future nobody wants to live in.”
That’s the shape of the current moment in medical coding: AI-assisted capture on the provider side, AI-assisted denial and downcoding review on the payer side, with the same underlying charts feeding both. Neither side is going to stand down first.
What This Means for Coding and Compliance Teams
Whatever share of the $942 million figure holds up to scrutiny, the operational message for hospital coding and compliance departments is unambiguous. A few things are worth acting on now rather than after the first payer audit letter arrives:
- Every AI-suggested diagnosis needs a documented clinical validation step before it reaches a claim — an audit trail showing a human reviewed and confirmed the supporting evidence, not just accepted a suggestion.
- Coding intensity metrics (case mix index trends, secondary-diagnosis capture rates) should be tracked internally before a payer’s own analytics flag an outlier pattern.
- CDI and coding teams need a shared standard for what counts as sufficient documentation to support a comorbidity code, since that standard is what payer-side AI models are increasingly built to test against.
- Denial-management workflows should assume payer AI review is now standard, not exceptional, and budget review capacity accordingly.
The Bigger Pattern: Trust Gaps on Both Sides
This dispute is really a preview of where Medicare Advantage risk-adjustment audits (RADV) and commercial payer reviews are both heading: less reliance on sampling, more reliance on AI models scoring every claim for coding-intensity anomalies. Hospitals that can show a clean, well-documented AI-assist workflow — one where every machine-suggested code has a traceable clinical basis — will be far better positioned than those that simply deployed a coding-assist tool and let volume rise without a matching validation process.
BCBSA’s finding also raises a fair question for AI-in-coding vendors generally: accuracy against a ground-truth chart is necessary but not sufficient if the industry outcome is an arms race that neither payers nor providers can defend to regulators. Coverage of the story, including reporting from Inc., frames it as a “bot war” — but the practical takeaway for coding departments is less dramatic and more procedural: documentation discipline is what will separate defensible AI-assisted coding from the pattern BCBSA just put a dollar figure on.
Getting Ahead of the Next Audit Cycle
None of this means AI-assisted coding is the wrong direction — the accuracy and consistency gains are real, and going back to purely manual coding isn’t a realistic answer to a payer dispute. What it means is that the tooling has to be paired with governance: validation steps, intensity monitoring, and documentation standards that hold up when a payer’s own AI comes looking. Coding teams that build that discipline in now will spend a lot less time defending individual claims later.
Medikode’s automated medical coding platform is built around exactly that pairing — automated code suggestion with a built-in clinical validation layer, so every AI-assisted code carries the documentation trail auditors and payers are increasingly going to ask for. Learn more at medikode.ai.