The AI DOGE Built to Cancel Veterans Contracts Couldn't Read Them Correctly
What happened
In June 2025, the Department of Government Efficiency deployed an AI tool to do something that would take a large team of human reviewers months: scan nearly 90,000 Department of Veterans Affairs contracts and identify which ones could be canceled. The speed was the point. What the tool produced instead was a series of errors significant enough to call the entire exercise into question.
The system, internally named MUNCHABLE and developed by DOGE, made fundamental mistakes when reading contract data. Among the documented failures, the tool misread contract values and inflated some of them dramatically, reporting figures in the tens of millions of dollars for contracts that were worth far less. An error of that kind does not just produce a wrong number. It corrupts the comparison used to decide whether a contract is wasteful, which means every cancellation recommendation derived from that misreading starts from a false premise.
The stakes were not abstract. The VA manages healthcare and services for millions of American veterans, and its contracts underpin the infrastructure of that care. Independent experts who reviewed the situation concluded that AI was the wrong technology for this task, not just because it made errors in this particular case, but because the work requires contextual judgment that large language models are not built to reliably perform on structured government procurement data.
The political fallout arrived quickly. Two US senators called for formal investigations into how AI was being used in the VA's contract review process, arguing that its deployment introduced unease around decision-making, security, governance, and quality control at every level. The employee credited with building the tool was fired after speaking publicly about his work, an outcome that illustrated how little accountability existed at the institutional level even as lawmakers pushed for more of it.
The incident is a clear example of what happens when an AI system is handed consequential authority before anyone has established how its outputs will be checked. There was no audit trail that allowed reviewers to trace which contracts had been flagged for the wrong reasons, no independent validation step before recommendations reached decision-makers, and no clear record of what the system actually processed. A provable record of what a system did, including what inputs it read, what it concluded, and who authorized the result, would not have prevented the underlying errors. It would have made them visible before they affected any veteran's care.
Reported impact
- Affected parties
- Not publicly disclosed
- Harm type
- Not publicly disclosed
- Scale
- Not publicly disclosed
- Financial impact
- Not publicly disclosed
- Regulatory action
- Not publicly disclosed
Classification
Relevant governance controls
Governance control mapping is not available for this record.
- No controls mapped
Not publicly disclosed
Control mapping is analytical. It does not state that any control would have prevented the incident.
Sources and evidence
This record was researched and written by the Index. The event is also catalogued in the following database, which is listed for cross-reference.