VA claims automation (automated survivor-benefit decisions)

The Department of Veterans Affairs (VA) automates survivor-benefit decisions for deaths connected to military service, called Dependency and Indemnity Compensation (DIC). Phased in from May 2020, the software reads scanned death certificates and applications and applies predefined rules. When every rule is met, it issues the rating decision, VA's formal ruling, plus the award and letter, with no human involvement.

How it works

VA's Veterans Benefits Administration (VBA) runs the automation. Its Pension and Fiduciary Service built it with VA's Office of Information and Technology. The VA Office of Inspector General (OIG), VA's independent watchdog, describes it as rules-based automation and document extraction, not machine learning.

Because the rules are predefined, a defect in them repeats on every claim it matches. VA says, in a figure the OIG report cites, that it makes more than 1,000 DIC payments or adjustments a day.

What the 2026 review found

The OIG issued its review on April 30, 2026, as report 25-00153-47. It covered automated DIC grants from September 2023 through August 2024.

At least 8,000 of an estimated 8,100 automated grants had at least one legal or procedural deficiency. The deficiencies included incomplete evidence summaries. Other decisions left out findings in the survivor's favor that the law requires a decision to state. Most rating decisions listed only the death certificate as evidence.

The money errors are a narrower count. At least 2 percent of the decisions, at least 190, had legal errors with a money impact. Together they came to at least $2.7 million, or $2,727,764 in questioned costs. That is the OIG's term for spending it found improper or unsupported.

They included death benefits granted for conditions unrelated to service without supporting medical evidence. They also included wrong effective dates, the dates from which payments start.

How the press reported it

Public Radio East, in June 2026, called it an automation "glitch" that "ruined 98%" of survivors' claims. That merges the two counts. The 98 percent counts any legal or procedural deficiency, not money errors.

Task & Purpose, in May 2026, reported one overpayment of $22,692 as an example. It quoted a Disabled American Veterans representative who supports automation only with robust human oversight. That representative's widely quoted figure of about 2,000 claims a day was a hypothetical, not a count.

How the problem was reported, and what changed

In April 2020, a VBA analyst reported through VA's internal defect-tracking system that automated DIC decisions listed only the death certificate as evidence. VBA's Pension and Fiduciary Service closed the report without action. The same deficiency was central to the 2026 findings.

In March 2025, right after the OIG's preliminary briefing, VA removed guidance from its manual that the sources call the long-form guidance. The sources do not say what it said. That change came about five years after the analyst's report. The OIG's full public report followed in 2026, about six years after it.

The OIG also found the quality-review checklist for automated DIC claims less rigorous than the review traditional claims receive. The checklist did not focus on legal and procedural requirements.

VA's modernization plan to Congress, under section 701(b) of the PACT Act, did not fully explain that VBA grants these claims end to end without human intervention.

While the errors continued

The VA Secretary announced expanded DIC automation in May 2025. As of November 2025, 20 more automated decisions from September and October 2025 showed similar errors.

The OIG's first recommendation, to strengthen the automation, remained open. VBA concurred with it only in part, and disputed part of the report's methodology.

Other automated tools in the same claims work

An OIG review in September 2023 (report 22-02936-175) looked at the hypertension automation of VBA's Automated Benefits Delivery program. Of 60 automated claims reviewed, from December 2021 to September 2022, 16 had inaccurate decisions. That is 27 percent.

Contractor-produced summary sheets lacked comprehensive blood pressure information. Under unclear guidance on which blood pressure counts as predominant, rating specialists left inaccurate extracted readings unresolved. The program's leaders told the OIG they did not know of the errors until the review.

A Special Monthly Compensation calculator gave wrong results for complex cases, such as limb loss, blindness, and aid and attendance. It sat inside the Veterans Benefits Management System for Rating (VBMS-R), the software VBA staff use to rate claims. In tested scenarios it underpaid $132 to $4,170 a month, and produced one overpayment of $373.

VBA and VA's IT office could not find the cause, or say how long the errors had gone on. After an allegation in November 2023, VA turned the calculator off in October 2024. It went back to a legacy calculator that works correctly. The review tested scenarios and did not identify affected veterans.

An April 2025 review (report 24-01153-52) looked at 131,000 claims from the first year of the PACT Act. About 31,400 of them, 24 percent, had incorrect effective dates. About 26,100 of those caused at least $6.8 million in improper payments, and $20.4 million was projected over three years. The OIG found two automated date tools unreliable.

The human work around the automation

The OIG documented errors in the human work around the automation too. In December 2024 it estimated that 45 percent of about 19,200 denied PACT Act presumptive claims had at least one error. Presumptive claims are for conditions the law presumes are connected to service. These were processors' errors under confusing guidance, not automation errors.

That review also estimated unwarranted medical exams in about 6,900 denials, costing about $1.4 million over six months. Two veterans were improperly denied and underpaid about $56,700.

In September 2025 the OIG reported on a senior benefits representative in Philadelphia, a staff member who authorizes decisions. She authorized about 85,300 claims in fiscal years 2022 to 2024, about 19 times the national average. She took 4.7 minutes per claim, against a norm of about 21 minutes.

In a sample of her authorizations from January to June 2024, about 84 percent had at least one error. The OIG estimated $2.2 million in improper payments from that window. Managers knew of her outlier volume but did not strengthen controls. The review also names production goal incentives.

In May 2026 the OIG found nearly 10,000 unwarranted or poorly justified overrides of software safeguards in VBMS-R in six months. They caused more than $67,000 in erroneous payments. There was no quality review or feedback process for overrides.

How the case file reads it

Several correction channels existed inside VA, and each one failed. The only channel that ever changed behavior was outside and slow: the OIG.

The automation decides claims continuously, while the outside audit arrives years apart. So errors built up between audits, and the program expanded while known errors persisted.

The automated grants recorded little evidence, most listing only the death certificate. Each such error made it less likely that the next reviewer, appeal, or audit would catch it.

Errors here run both ways. Survivors were underpaid and overpaid, and effective dates were set too early and too late.

Where the facts come from

The facts come from six OIG reports, from 2023 to 2026. They cover the DIC automation, the claims automation project, the Special Monthly Compensation calculator, PACT Act effective dates, PACT Act denials, and one senior benefits representative's authorizations. Reporting by Task & Purpose, Public Radio East, Stars and Stripes, Military.com, and FedScoop adds detail, including FedScoop's account of the 2026 override review.

This example follows the shape of these tools. It does not rebuild them.

What the available tools can and cannot address

A failure pathway is a link between two parts of the network where a mistake made by one part can be passed on to the other. A tool closes a pathway when mistakes stop passing along it. The work along it may go on.

This case has a budget of 12 units. Explore (No Targets) sets no targets. Contained means the network's mistakes are corrected rather than building on each other. The Lab shows the network's state as self-correcting, at a tipping point between the two, or in cascading failure, where mistakes keep building on each other.

The Lab starts at the setting Both, with side effects and lingering effects both on. Under Explore and Service Targets Only, at that setting, two tools costing 4 of the 12 units keep the mistakes contained. Under Service Targets Only the same pair also meets the service target. That target asks for the automated system to be helping the work.

One such pair is Mark AI-written records with Peer sharing rules. Another is Escalate checks with Mark AI-written records. Here Mark AI-written records marks the automation's entries in the benefit record, so processors and the automation reading them can weigh them.

With the setting at Off, where both effects are off, Vet connections alone does the same, for 3 units. At the starting setting it works at reduced strength unless Understand the system is also on.

Under Service and Safety Targets the targets can be met. That level adds closing every failure pathway, and it asks that the work not be strained. The cheapest combinations cost 10 of the 12 units and use four tools.

One is Escalate checks, Understand the system, Vet connections, and Store less data. Understand the system costs 4 units at this level. While it is on, Vet connections and Store less data cost 2 units each instead of 3.

Counting stronger settings, 17 distinct sets of tools meet these targets. Every one includes Escalate checks and Vet connections.

Under All Governance Targets the targets can be met too. That level asks for the automated system to be clearly helping the work. Escalate checks, Understand the system, Vet connections, and Store less data meet them for 10 units. All 6 distinct sets of tools that meet them include those four.

More is not better here. Every tool at its stronger setting at once costs 35 units, far over the budget. It contains the mistakes and closes every failure pathway, but leaves the automated system hurting the work. So it misses the targets at every level that sets them.

At the starting setting no single tool contains the mistakes. Used alone, Upgrade model, Keep prompts neutral, and Check copied records leave the network in cascading failure. So do Review on schedule and Store less data at their standard settings. No tool offered here adds an independent second check on each automated decision.