Flawed AI Tool Used by DOGE to Review Veterans Affairs Contracts Raises Concerns

DOGE's AI Tool for VA Contract Review

The Department of Government Efficiency (DOGE) recently employed an artificial intelligence tool to review and potentially cancel contracts at the Department of Veterans Affairs (VA). This initiative, part of the Trump administration's efforts to streamline operations and cut costs, has raised significant concerns due to the tool's flaws and potential impact on veterans' healthcare 1

1

2

2

3

3

.

Development and Implementation of the AI Tool

Sahil Lavingia, a software engineer with no prior healthcare or government experience, was tasked with creating the AI tool. Working under tight time constraints, Lavingia developed the tool in just two days, using outdated and inexpensive AI models 1

1

. The tool was designed to identify "MUNCHABLE" contracts – those deemed non-essential and potentially cancelable 1

1

3

3

.

Flaws and Errors in the AI System

The AI tool exhibited several critical flaws:

1. Contract value hallucinations: The system frequently misread contract values, inflating them significantly. For instance, it erroneously concluded that over a thousand contracts were worth $34 million each, when some were actually valued at just $35,000 1

   1

   3

   3

   .

2. Limited input processing: The AI only analyzed the first 10,000 characters (approximately 2,500 words) of each contract, potentially missing crucial information 2

   2

   .

3. Lack of context: The tool lacked a sophisticated understanding of medical care, institutional management, and human resources necessary for accurate decision-making 2

   2

   .

4. Unclear instructions: The prompts given to the AI model were often vague or conflicting, leading to misinterpretations 2

   2

   .

Impact on Veterans Affairs Contracts

The DOGE AI tool flagged over 2,000 contracts for "munching" or potential cancellation 1

1

3

3

. While the exact number of canceled contracts remains unclear, VA officials have reported terminating nearly 600 contracts overall 1

1

3

3

. Some of the canceled contracts included:

• Maintenance of gene sequencing devices for cancer treatment research
• Blood sample analysis for VA research projects
• Tools to measure and improve nursing care 1

  1

  3

  3

Expert Opinions and Concerns

AI and procurement experts who reviewed the code and prompts expressed significant concerns:

• Cary Coglianese, a professor at the University of Pennsylvania, criticized the use of general-purpose large language models for such complex tasks 1

  1

  3

  3

  .
• Experts noted that using off-the-shelf AI models with little context on VA operations was problematic 2

  2

  .
• The rushed implementation and lack of proper tools were cited as contributing factors to the system's flaws 2

  2

  .

VA's Response and Future Implications

VA press secretary Pete Kasperowicz defended the contract review process, stating that decisions to cancel or reduce contracts undergo multiple reviews by VA employees, including contracting experts and senior staff 1

1

3

3

. However, the VA's plan to simultaneously move services in-house while slashing staff has raised questions about the feasibility of maintaining quality care for veterans 1

1

.

As the Trump administration aims to cut around 80,000 people from the VA's workforce of nearly 500,000, concerns grow about the potential impact on veterans' healthcare services 1

1

3

3

. The use of flawed AI tools in such critical decision-making processes highlights the need for careful consideration and expertise when implementing AI in government operations.