On September 8, 2026, President Donald Trump signed Executive Order 14426. The order, published in the Federal Register three days later, directs the Department of War and the Department of Veterans Affairs to improve the way military personnel files and service treatment records move between them. It also directs the agencies to create digital tools that use artificial intelligence and other technologies to simplify veterans' applications for benefits.

Both jobs have the same 180-day deadline. A separate 30-day provision says current systems should share a service member's records with VA immediately upon discharge or release. A 120-day provision requires a review of relevant technology contracts for interoperability.

Those are agency instructions, not a new claims portal. The order names no product or vendor, and it does not decide a single veteran's claim. It is subject to existing law and available funding, and it creates no private right to a faster decision.

It still matters because the application and the underlying record are being rebuilt on the same clock. A cleaner form can arrive before every medical or service record behind it is easy to find.

What VA already has in motion

VA was using automation in benefits work before this order. Its public AI inventory lists hundreds of use cases across the department. Some handle narrow tasks such as finding documents or prepopulating parts of a toxic-exposure memo. Many benefits-processing projects remain in development or pre-deployment.

One of them is the Artificial Intelligence Claims Evaluation System, or AICES. An April 2026 VA privacy assessment describes AICES as an evaluation of whether AI can help triage disability claims for possible Acceptable Clinical Evidence review. That process may allow a clinician to review existing records instead of requiring an in-person examination.

The assessment says AICES can read structured and unstructured VA data, use optical character recognition and natural-language tools, and create a Disability Benefits Questionnaire from records already in VA systems. A successful questionnaire can then be uploaded into the Veterans Benefits Management System, where a government employee can use it in a claim decision. VA says the project could reduce manual work, processing time, and unnecessary travel.

That description is narrower than the claim that a robot will decide every case. AICES is an evaluation, not evidence of a nationwide rollout. The public record used for this article does not show that the September executive order will use AICES at all.

At a July House hearing, VA acting deputy chief information officer Robert Orifici stated the agency's policy plainly: "Every disability claim is decided by a trained VA employee, not by AI or automation. These tools support human decision making; they do not replace it."

A human decision is a meaningful safeguard. It does not answer the next question: What did the human receive?

A signature cannot restore a missing record

A claims processor may see an original document, a machine-produced summary, a derived form, or some combination of them. If a medical record was never transferred, a statement was tagged incorrectly, or a summary omitted a detail, the human can make a careful decision from an incomplete file.

The VA Office of Inspector General found a version of that problem in an automated survivor-benefit workflow. Its April 30 report examined automated decisions granting service-connected death benefits from September 2023 through August 2024, then checked additional cases through November 2025.

Of an estimated 8,100 decisions or notification letters in the main review period, at least 8,000 had one or more legal or procedural deficiencies. Common problems included incomplete evidence summaries and omitted favorable findings. The inspector general also estimated that at least 2% of decisions contained legal errors that led to at least $2.7 million in improper payments. Its review found that automated decisions received less rigorous quality checks than traditionally processed claims.

The report explains why the evidence list is more than paperwork: "when VA does not clearly document the evidence considered in a rating decision, claimants cannot be certain that all submitted materials were reviewed."

That review concerned one type of survivor claim and a specific automated process. It was not an audit of every disability claim. It also does not prove that generative AI caused the problems. The system applied predefined rules without human involvement, according to the inspector general. Mixing that finding with every newer AI project would turn a useful warning into a vague accusation.

The narrower lesson is stronger. Automation can move a file faster while weakening the record of what the file contained.

What the evidence cannot tell us

The executive order leaves the specific tools, testing methods, and missing-document handling unspecified. Public records do not identify whether any particular claim received AI assistance or provide an error rate for VA claims as a whole. They also contain no completed implementation record for this order.

The inspector general's figures also do not mean that a missing item in a decision letter changed the outcome. An omission may be serious, harmless, or the result of a document arriving somewhere else in the file. A veteran needs the actual claim record and qualified help to sort that out.

What the records do show is a gap between two assurances. VA says a trained employee makes each disability decision. Oversight found that an automated workflow sometimes gave claimants an incomplete account of the evidence used. Both statements can be true.

Keep a small evidence receipt

When you submit material for a VA claim, keep a plain list of what you sent, when you sent it, and how. Save the upload confirmation or mailing receipt with that list. Use filenames or short descriptions you will recognize later, such as the clinic name and date, instead of writing only "medical records."

When a decision letter arrives, find the section that lists the evidence considered and compare it with your list. Ask one question: Does the decision letter name the same evidence you submitted?

A mismatch does not prove that VA made the wrong decision. It gives you a specific issue to raise instead of a general fear about AI. VA says accredited representatives and employees at a regional office can help with claims, and many Veterans Service Organizations provide accredited help at no charge. Ask what is in the claim file and what review route, if any, fits your situation.

This is a records check, not legal, benefits, medical, or privacy advice. The systems may get faster. Keep the evidence list anyway.

Public sources