The Machine Signed Off, But A Person Gets Blamed.
There’s a number in this story that needs to be addressed.
Eight thousand one hundred death benefit decisions. The VA’s own inspector general looked at them. Eight thousand had a significant error.
That’s ninety-eight percent.
Not ninety-eight percent perfect. Ninety-eight percent wrong, or wrong enough that the watchdog felt the need to flag it. These aren’t parking tickets. These are death benefits — the money a family gets after they’ve already lost someone who served.
An attorney named Benjamin Krause has spent years representing veterans in front of the VA. He sat down for an interview this week and laid out exactly where the trust breaks down. And it isn’t where you’d expect.
The VA’s defense is simple. They say a human signs off on the final decision. AI might touch the process along the way, but a person makes the last call. Case closed, they say. Nothing to see here.
Krause says that’s not good enough. And once you hear his reasoning, it’s hard to argue with him.
Think about what actually happens before a veteran gets a letter in the mail. Evidence gets weighed. Files get reviewed. A decision on whether to send someone for a medical exam gets made. Notes get written up, sometimes by AI listening in on an appointment. All of that happens before any human looks at a final page and stamps it approved.
If AI shaped any of those steps — and Krause says it clearly does — then a human signature on the last page doesn’t mean much. The bias, the error, the misreading of evidence, all of that is already baked in by the time a person even sees it. The human isn’t checking the machine’s work. The human is rubber-stamping it.
That’s the real problem. Not that AI is being used. It’s that the VA is disclosing the smallest possible piece of where AI touches the process, and calling that transparency.
Imagine hiring a contractor to build your house. He tells you a licensed inspector signs off on the final walkthrough, so don’t worry about anything else. But the framing, the wiring, the foundation — all of that was done by someone else entirely, someone you never met, using methods nobody explained to you. Would you feel safe just because a name is on the last piece of paper?
That’s what’s happening to veterans right now.
Krause also flagged something just as troubling. The VA doesn’t even like the word “AI.” Dig through their reports and you’ll find the word “automation” instead. Same technology, softer name. It’s the kind of word choice that keeps Congress from asking the right questions, because on paper, it doesn’t sound like the thing everyone’s worried about.
Words matter. If you can’t call a thing what it is, you can’t be honest about what it does.
Now, to be fair, there’s a real upside here. If the VA gets this right, claims could move faster. Backlogs could shrink. Veterans waiting months, sometimes years, for a decision could get answers sooner. Nobody wants to stand in the way of that. Speed matters when someone is struggling to pay rent while they wait on a claim.
But speed built on a ninety-eight percent error rate isn’t progress. It’s just moving the mistake faster.
This is the whole argument in one sentence: a rule that only requires disclosure at the very last step is a rule that technically gets followed while completely missing the point. The VA can say they’re being transparent and still leave veterans with almost no idea how their case was actually handled. Checking a box isn’t the same as earning trust.
Real transparency would mean flagging AI’s fingerprints everywhere they show up — not just the final signature, but every step that led there. It would mean telling veterans plainly when a machine helped weigh their evidence, or wrote up their appointment notes, or flagged them for a certain kind of exam. Not because the technology is evil. Because people deserve to know who, or what, is deciding their future.
Veterans didn’t sign up to have a computer’s judgment hidden behind a human’s signature. They signed up trusting that whoever handled their case would tell them the truth about how it was handled.
Right now, that trust is not being earned. It’s being assumed.
This post was drafted with AI governed assistance and reviewed and directed by Michael S. Faust Sr. before publication.
Contact: micvicfaust@gmail.com
© 2026 The Faust Baseline LLC | All Rights Reserved
