DOGE plans for AI government alarm experts on rights & security
DOGE has fed sensitive personal & financial information into AI systems, which transfer data to private servers; plans for AI decision-making in government create grave risks to rights & security.
Reports are emerging that Elon Musk’s so-called ‘Department of Government Efficiency’ (DOGE) is planning to install AI systems to manage public services now run by trained, oath-taking civil servants. Experts with knowledge of related data security, national security, rule of law, and rights protections, are increasingly alarmed that vital services may be handed over to unreliable “chatbot” systems which, at best, can only guess at appropriate decisions.
DOGE is an informal advisory panel that has been acting as a decision-making layer of the Executive branch, in principle reporting to the White House Chief of Staff. There have been few public indications that anyone other than Elon Musk is in charge of the DOGE operation, and a strange press conference earlier this week, in which Elon Musk stood behind the Resolute Desk and spoke in place of President Trump, only confirmed that suspicion.
Here, we offer a brief rundown of reporting and analysis regarding the alleged plans of Musk and DOGE to replace sworn civil servants with AI. We include a couple of points from each, for easy reference.
WIRED: DOGE Has Started Gutting a Key US Technology Agency
“At least dozens of workers for the Technology Transformation Services, housed within the US General Services Administration, were fired Wednesday afternoon, sources tell WIRED.”
“GSA’s vision for the TTS, under new leadership, is an agency that runs like a “startup software company,” WIRED reported last week, focusing on AI, automating different internal processes, and centralizing data from across the federal government.”
The Atlantic: Elon Musk Is Breaking the National-Security System
“To protect secrets, people who will be handling classified information or assuming positions of trust within intelligence agencies are vetted, often by law-enforcement agents, who interview friends and co-workers, review travel histories, and analyze financial information to determine whether someone might make an attractive recruit for a foreign intelligence service.”
“The fact that people are getting access to classified and personally identifiable information who are not being vetted by our national-security system means it is more likely that there are going to be damaging leaks.”
WIRED: Elon Musk’s DOGE Is Working on a Custom Chatbot Called GSAi
“ELON MUSK’S DEPARTMENT of Government Efficiency (DOGE) is pushing to rapidly develop “GSAi,” a custom generative AI chatbot for the US General Services Administration…”
“One tool the team looked into, according to documents viewed by WIRED, was Cursor, a coding assistant developed by Anysphere, a fast-growing San Francisco startup. Anysphere’s leading investors include Thrive Capital and Andreessen Horowitz—both of which have connections to Trump.”
Bulletin of the Atomic Scientists: Why DOGE’s meddling at Treasury could have catastrophic consequences for the US economy
“Individual engineers experimenting directly upon real, live systems without first validating their changes or seeking peer review violates every best practice for controlling risk in the industry.”
“…it’s only a matter of time until DOGE’s meddling inadvertently triggers a catastrophic failure of Bureau of the Fiscal Service systems, and the damage may not be reversible, if the safeguards required to run a secure, reliable system have been bypassed.”
The Washington Post: Elon Musk’s DOGE is feeding sensitive federal data into AI to target cuts
“Feeding sensitive data into AI software puts it into the possession of a system’s operator, increasing the chances it will be leaked or swept up in cyberattacks. AI can also make errors, for example hallucinating incorrect information when summarizing data.”
“…according to two people with knowledge of the DOGE team’s actions… The AI probe includes data with personally identifiable information for people who manage grants, as well as sensitive internal financial data…”
Bloomberg: Dismissed Nuclear Bomb Specialists Recalled by Energy Department
“The employees, responsible for designing and maintaining the nation’s cache of nuclear weapons at the National Nuclear Safety Administration, were part of a larger wave of workers dismissed from the Energy Department, drawing alarm from national security experts.”
“The firings — part of a wave of terminations across the federal government this week spurred by Elon Musk’s Department of Government Efficiency — underscore the chaos as the world’s richest man seeks to quickly overhaul the federal bureaucracy in Trump’s image.”
There are widespread concerns about how DOGE is operating inside of sensitive U.S. Government systems and reports of unlimited direct access to all of the personal information the Department of Treasury and other agencies have about American citizens and businesses. One particularly glaring security concern, even if nothing nefarious were done, is the reported inputting of massive amounts of sensitive data into generative AI systems.
Such systems need to be able to read everything fed into them as inputs, so they can provide an approximately relevant response. They also tend to store information, often without provenance, which means they might regenerate that information later in an apparently unrelated output. ChatGPT, for instance, has been shown to be able to be tricked into generating plagiarized and/or identifying information.
The Leadership Conference on Civil and Human Rights—a coalition of more than 240 organizations—has called for an immediate halt to DOGE access to sensitive data, citing potential irreversible harm to the privacy and data security of individual Americans, as well as businesses and institutions.
A letter from the rights coalition reads, in part:
“This unchecked breach of our most sensitive data is shocking and illegal.[2]Musk and his employees and agents are not elected officials, do not appear to have been vetted for security clearances, or were not even government employees at the time they accessed government systems and the personal data within them.[3] Moreover, they are threatening individuals who merely seek to identify who is participating in these violations.[4]”
 | A guest post by
|