The Validation Report Nobody Reads at 23:00

Labcorp announced the nationwide rollout of PathAI's AISight Dx digital pathology platform on 23 February 2026. The press release positioned it as FDA-cleared infrastructure for fully digital workflows across anatomic pathology labs and hospital collaborations, with the chief scientific officer promising faster, more consistent results. What the announcement did not mention is the 60-case validation set required before a single diagnostic slide goes live, the LIS integration that fails in 57% of surveyed labs, or the person who answers the page when the scanner stops mid-shift and a frozen section is waiting.

That person is not celebrating the AI era. She is deciding whether to call the vendor at 23:00 or revert to glass, and the decision has a 12-minute window before the surgeon closes without a margin call.

The FDA has authorized over 1,600 AI-enabled medical devices for marketing in the United States as of September 2026, with radiology accounting for the overwhelming majority. Pathology has lagged, not because the models are worse but because the deployment surface is different. A radiologist triages a worklist; a pathologist signs out a diagnosis. The College of American Pathologists recommends a validation set of at least 60 cases, with consistency between whole-slide images and glass slide reads greater than 95%. That is the entry cost, and it is paid in pathologist time before the platform does a single useful thing.

The validation is not a formality. It is the contract that makes the lab manager personally liable when the AI-assisted workflow produces a discordant result six months later. In 2024, CAP estimated that digital pathology adoption is approximately 10% in U.S. labs. The other 90% are not waiting for better models. They are waiting for someone else to absorb the operational risk of going first, and they are correct to wait. An international survey conducted in 2023 across 127 laboratories reported that 57% had implemented digital pathology for research or clinical purposes, citing improved turnaround time, case traceability, and multi-site collaboration, but the survey also reported challenges with laboratory information systems integration and a lack of workforce training.

LIS integration is the politest way to describe the failure mode. The whole-slide image leaves the scanner as a DICOM or proprietary format file, typically 2-4 GB for a single H&E slide at 40x magnification. It must flow into the LIS, attach to the correct accession, be retrievable by the pathologist at the correct workstation, and remain accessible for the legally mandated retention period. At 200 or more slides per day, storage becomes one of the primary ongoing operating costs, and labs must define the storage policy (cloud, on-premise, or hybrid) before going live. The press release does not say who pays for three years of cloud storage when the hospital IT budget was set 18 months ago, or what happens when the vendor's SDK update breaks the LIS connector and cases sit in a queue with no attached metadata.

The lab manager pays. She pays in escalations, in manual workarounds, and in the meeting where she explains to the medical director why turnaround time spiked 40% in the first month after go-live. Personal opposition to digital pathology was cited as the main reason for low adherence in 41.7% of surveyed labs, followed by concerns over slower turnaround time for digitized cases in 31.9%. That opposition is not Luddism. It is pattern recognition. The pathologist who refuses to sign out digitally has seen what happens when the scanner backlog hits 90 slides at 16:00 and the evening shift has four people instead of six.

Labcorp's deployment carries a different risk profile than a single-site academic rollout. The platform will be deployed across anatomic pathology labs and hospital collaborations, enabling fully digital workflows for case management, slide review, collaboration, and annotation, with AI-powered image analysis, secure storage, and system connectivity. That is a correct description of the technical capability and an incomplete description of what it means to operate it at scale. A national network means that a scanner failure in one site can cascade into workflow disruptions across three hospitals if the contingency plan assumes centralized digital review. The AI-powered image analysis is a feature when it works and a liability when it flags a benign lymph node as suspicious and the pathologist has to document why she overrode the model.

Most clinical implementations follow a human-in-the-loop model, where AI-generated outputs are reviewed and validated by pathologists before final diagnosis, ensuring clinical accountability and minimizing automation bias, but key challenges include model generalizability across diverse populations, variability in slide preparation, and data drift over time. Data drift is the term of art for what happens when the model was trained on one lab's staining protocol and deployed in another's. The AI does not know that the hematoxylin batch changed in March, or that the new histotech cuts 5-micron sections instead of 4. It flags differently. The pathologist sees the flags, questions the model, and returns to glass to confirm. The efficiency gain disappears. The capital cost does not.

Labcorp's expansion builds on a 2019 strategic investment in PathAI, a collaboration that has since advanced to include AI-driven clinical trial support and validation of novel AI-pathology solutions. The clinical trial work is where the business model makes sense: controlled cohorts, centralized review, and a sponsor paying for compute and storage. Community hospital pathology is a different trade. The volume is unpredictable, the case mix is wide, and the payer reimburses the same rate whether the slide was reviewed digitally or on glass. CAP advocated for the implementation of 30 new Category 3 CPT codes in 2024 to highlight the additional work involved in digitizing surgical pathology slides for primary diagnosis, but advocacy does not automatically translate into reimbursement. Until it does, the lab absorbs the cost as overhead and the CFO asks why the digital pathology line item grew 22% year-on-year while the slide count grew 6%.

The pathology staffing shortage is real, and digital workflows are one of the few interventions that can let a three-pathologist group cover five sites. In an analog environment, where diagnoses rely on physical slide transport, manual workflows, and localized staffing, there is little flexibility to absorb shocks; without digital pathology, staffing shortages become productivity crises, turnaround times lengthen as volume grows, and access to subspecialists remains constrained. That is the case for deployment, and it is sound. The error is treating deployment as a binary event rather than a multi-year process of workflow redesign, technical troubleshooting, and staff retraining. Staff training, workflow redesign, and validation protocols all consume time from people already working at capacity. The vendor provides a two-day onboarding workshop. The lab manager provides the next six months of daily troubleshooting, because the workshop did not cover what to do when the scanner jams on a thick GI biopsy or the monitor calibration drifts and colours shift.

Demonstrating diagnostic concordance between digital platforms and traditional microscopy is essential for regulatory compliance and for developing user trust; following CAP digital pathology validation guidelines offers a reliable approach, with validating a representative set of cases through both digital and glass slide review helping to confirm diagnostic accuracy. The validation guideline is written as if labs have spare pathologist time to read 60 cases twice, document the concordance, and file the report before go-live. In practice, the validation happens in the margins: evenings, weekends, and the week between Christmas and New Year when elective case volume drops. The lab manager schedules it, tracks it, and escalates when it stalls because one pathologist is on leave and another refuses to participate until the monitor situation is resolved.

The monitor situation is that digital pathology at diagnostic resolution requires a calibrated, high-brightness display, and the standard-issue hospital IT monitor is neither. Respondents reported difficulty in recognizing fine details on digital slides, such as microorganisms, in 52.8% of cases. A microorganism missed on a digital slide is a patient harm event and a lawsuit. The lab manager cannot deploy until every workstation meets spec, and the capital request for six medical-grade displays sits in the queue behind the CT replacement and the EMR upgrade. She proceeds anyway, using the monitors she has, and documents the risk in an email that will be cited in the root-cause analysis if something goes wrong.

None of this is hidden. Transitioning from glass slides to whole-slide imaging can disrupt well-established processes, particularly in high-volume laboratories or systems operating across multiple sites; a phased rollout allows teams to adapt gradually while minimizing operational friction, and many institutions benefit from beginning with targeted use cases such as tumor boards, frozen sections, or consultative cases. The phased rollout is the correct answer, and it is the one that does not make the press release. Labcorp's announcement describes a nationwide platform. The lab manager reads that and knows it means her site goes live in Q3, she has 90 days to validate, train, and integrate, and the go-live date will not move because the contract is signed and the vendor's revenue forecast depends on it.

The AI component adds a second layer of dependency. AI diagnostic algorithms don't exist in isolation; every tool that grades prostate cancer, quantifies HER2 expression, or flags regions of interest depends on two things: whole slide images and an image management platform to integrate through. The image management platform is the vendor's software, the AI tool is a third party's model, and the LIS is the hospital's legacy system. When the AI stops flagging correctly, the lab manager has three vendors to call and no service-level agreement that assigns accountability across the stack. She calls all three, opens a ticket with each, and waits. While she waits, the pathologist reverts to manual review, the efficiency case for digital evaporates, and the CFO asks whether the ROI model was based on actual data or aspirational assumptions.

The assumption was that digital pathology would reduce sign-out time. Implementation improves turnaround time, case traceability within departments, and multi-site collaboration, but that is the optimized steady state, not the first six months. In the first six months, turnaround time increases because pathologists are slower on the new interface, the scanner queue is not yet tuned, and every edge case that used to take 30 seconds on glass now takes three minutes of navigation, zooming, and second-guessing. The lab manager tracks the metrics, sees the spike, and writes the explanation that goes into the quarterly quality report. The explanation is accurate and irrelevant. The metric is red, the hospital benchmarks against peer institutions that have not deployed digital, and the lab looks slower than it was.

By month eight, if the deployment does not collapse, the curve inverts. Pathologists are faster digitally than they were on glass, because they can pull up comparison slides instantly, annotate regions for the fellow to review, and consult with a subspecialist in another time zone without shipping a box of slides. The lab manager stops getting paged about scanner jams because the histotechs have learned to stage slides correctly. The storage cost is still growing, but it is predictable, and the IT team has accepted it as a permanent line item. The validation report that nobody read at 23:00 on go-live night is now filed in the accreditation binder, and the CAP inspector will glance at it for 45 seconds during the next survey.

That is the successful case. The lab manager who gets there has absorbed a year of operational risk that the press release did not mention and the vendor's sales deck did not model. She did it because the alternative was worse: a staffing cliff with no digital infrastructure to distribute the load, and a five-year horizon where every peer institution has deployed and hers has not. The announcement said the technology allows labs to scale digital pathology nationwide and integrate AI insights into routine care. The lab manager knows it allows labs to attempt to scale, conditional on validation, integration, training, storage, workflow redesign, and vendor reliability. She also knows that when the system breaks at 23:00, the announcement will not answer the page. She will.

Tarry Singh is the founder and CEO of Real AI (realai.eu), an enterprise AI advisory and deployment firm working with global enterprises on production agent systems, model risk, and AI sovereignty strategy. He also leads Earthscan (earthscan.io) for Energy AI, and is a founding contributor to the EU-funded HCAIM and PANORAIMA programmes for responsible AI education across European universities. He writes at tarrysingh.com.