By Key Points
- DOGE AI Tool Mistake Cancels 585 Vital Veteran Affairs Contracts
- Flawed AI used by DOGE read only first 10,000 characters of contracts
- 585 VA contracts worth $900 million were canceled due to AI errors
- Critical terms like “core medical” and “DEI” were undefined in code
- AI hallucinated contract amounts and misunderstood key priorities
The Department of Government Efficiency (DOGE) is under fire after using a flawed AI tool that mistakenly flagged hundreds of Veterans Affairs (VA) contracts for cancellation.
In just 30 days, DOGE engineers built an AI system to comply with President Donald Trump’s executive order aimed at eliminating unnecessary government spending. But this rush led to serious problems.
Engineer Sahil Lavingia designed the tool to “munch”—or cancel—contracts not seen as directly supporting patient care. However, the AI was only set to read the first 10,000 characters of each contract, missing crucial details buried deeper in the documents.
How DOGE Created an Error-Prone AI Program to Tear Up Veterans Affairs Contracts https://t.co/pgoZyMkM3C
— #TuckFrump (@realTuckFrumper) June 6, 2025
Experts say the AI also suffered from poorly defined instructions. It was told to eliminate diversity, equity, and inclusion (DEI) programs but was not given a clear definition of what DEI meant. Similarly, it was asked to assess if pricing was “reasonable,” without any benchmark for reasonableness.
To make matters worse, the AI was built on an older general-purpose model not designed for complex government procurement tasks. This led the system to “hallucinate,” inventing incorrect contract amounts and misinterpreting their importance—an issue not uncommon in some AI models, as seen in OpenAI’s recent efforts to improve AI reliability through its deep research AI agent.
The VA “may deploy #AI to help the agency replace employees. Documents previously obtained by ProPublica show DOGE officials proposed in March consolidating the benefits claims department by relying more on AI”: https://t.co/xjgKylvOoS #ethics #tech #gov
— Internet Ethics (@IEthics) June 6, 2025
Despite these flaws, the tool flagged hundreds of contracts. Initially, the VA announced the cancellation of 875 contracts. But after pushback from veteran advocates and internal reviews, that number was revised to 585 non-critical or duplicate contracts, saving about $900 million.
3/ Second, the DOGE AI tool’s underlying instructions were deeply flawed.
The system was programmed to make intricate judgments based on the first few pages of each VA contract — about 2,500 words — which contain only sparse summary information. pic.twitter.com/Ix1rF57kyl
— Brandon Roberts (@bxroberts) June 6, 2025
Experts slam DOGE AI tool for lack of sophistication and oversight
DOGE’s reliance on a rushed AI solution has sparked widespread concern. According to Cary Coglianese, a University of Pennsylvania professor specializing in government AI use, determining which VA services could be handled internally requires “sophisticated understanding of medical care, institutional management, and human resources.” None of these elements were factored into the tool.
The code reviewed by ProPublica revealed more troubling signs. It contained vague terms like “core medical/benefits,” which the AI couldn’t meaningfully interpret. The model also failed to differentiate between contracts essential for safety inspections, veteran communications, and doctor recruitment—key functions necessary for veteran care.
A Department of Government Efficiency employee created a rushed AI tool to help uncover ‘waste, fraud, and abuse’ (Getty)
Further complicating the situation, many of the engineers working on the AI project were recruited from outside government circles. Lacking institutional knowledge, they were ill-equipped to understand the nuances of VA operations.
Interestingly, this is not an isolated case of AI being used prematurely in sensitive domains. Similar concerns were raised when Google’s Gemini updates showcased revolutionary AI capabilities that still required significant oversight and refinement.
One source told Federal News Network that the resulting contract cuts reflected a “communication breakdown” between DOGE advisors, VA leadership, and lawmakers overseeing the agency.
Lavingia admitted that the AI had limitations and that “mistakes were made.” He also insisted that all flagged contracts underwent additional human review before cancellation. However, critics argue this vetting process was inadequate given the AI’s serious flaws.
Musk hired engineers outside of the government to assist in finding ‘waste fraud and abuse’ in the government – as a result, many do not possess necessary institutional knowledge (AP)
Veteran advocates sound the alarm over essential services lost
Veterans’ groups quickly raised red flags after the initial cancellation announcement. Many of the 875 contracts included programs that ensured safe and effective medical facility operations and maintained critical communications between the VA and veterans.
Among the contracts initially flagged were those related to building safety inspections, communications about veterans’ benefits, and recruiting new doctors—functions vital to the VA’s mission.
After hearing concerns from advocacy groups and internal VA teams, the department walked back its cuts, revising the cancellation list to 585 contracts and stressing that only “non-mission-critical or duplicative” agreements were affected.
Still, veteran advocates remain wary. Many argue that such rushed AI-driven decisions could erode trust in the system designed to care for those who served the country. They point to broader concerns in the AI space, as even AI leaders like OpenAI CEO Sam Altman acknowledge that AI can behave in unexpected ways—often costing millions in errors and unintended consequences.
With ongoing advances like the Gemini 2.5 Pro update pushing AI boundaries, experts stress the need for robust human oversight before relying on such tools in critical government functions.
What's your reaction?
