The Trump administration has acknowledged that some members of Elon Musk’s Department of Government Efficiency may have improperly accessed and shared Americans’ Social Security numbers to assist a political advocacy group in efforts to overturn election results in certain states. Court documents reveal that in March 2025, these members were approached by the advocacy group to analyze state voter rolls with the stated aim of uncovering voter fraud. A DOGE team member signed a “Voter Data Agreement” with the advocacy group, potentially leading to violations of the Hatch Act, which prohibits federal employees from engaging in political activities using their official positions. A federal judge had previously blocked DOGE’s access to Social Security Administration systems due to concerns over privacy and data misuse, highlighting ongoing issues regarding the protection of sensitive personal information.
A new fraud-detection AI model, developed by the U.S. government’s Pandemic Response Accountability Committee, could have identified tens of billions of dollars in potentially fraudulent COVID-19 loan payments before they were disbursed. During a recent congressional hearing, PRAC executive director Ken Dieffenbach stated that the AI, trained on approximately five million Economic Injury Disaster Loan applications, could process 20,000 applications per second, highlighting its capability to detect anomalies and patterns consistent with fraud. The ongoing project aims to extend beyond COVID-19 fraud, with a mandate and funding provided to enhance oversight across various government programs.
Why do we care?
AI doesn’t break trust. People with unchecked authority do.
The DOGE situation wasn’t about clever technology. It was about privileged access being repurposed outside its mandate. That’s not an AI lesson—that’s a governance failure with a tech wrapper. This is authority governance: who gets access, under what mandate, and with what consequences. And if you think automation would have made that safer, you’re kidding yourself. It would have made it faster.
Now look at the fraud-detection model. That’s AI done right. It didn’t replace humans. It gave them a better flashlight earlier in the process. The tragedy isn’t that fraud happened—it’s that the insight existed but wasn’t operationalized soon enough.
For MSPs, the same pattern is playing out in your world. If you automate actions without redefining authority, you’re scaling liability. If you use AI to surface insights and keep humans accountable, you’re reducing risk.
Deploying AI is not as a shortcut to responsibility. The smart move is using it as a force multiplier for judgment.
This isn’t about trusting machines or distrusting people. It’s about being honest about who owns the consequences when systems act.