homehome Home chatchat Notifications


Federal Workers Say They’re Being Watched by AI for Saying Anything Bad about Trump or Musk

AI monitors federal workers for ‘anti-Trump’ and 'anti-Musk' language as oversight erodes, insiders say.

Tibi Puiu
April 8, 2025 @ 11:42 pm

share Share

Credit: FMT/EPA.

In early January, a peculiar warning rippled through the offices of the Environmental Protection Agency. It didn’t come in an official memo or policy update. Instead, it was whispered among managers, passed down from political appointees to federal employees: Be careful what you say, what you type and what you do.

The warning referred not to a new code of conduct, but to a new kind of surveillance — one allegedly powered by artificial intelligence and quietly embedded by a team working directly under Elon Musk.

The Department of Government Efficiency (DOGE) has become one of the most controversial instruments of the Trump administration’s second term. It was launched with the stated goal of slashing $1 trillion in federal spending and eliminating “waste, fraud and abuse.” Since then, DOGE has been rapidly redrawing the boundaries of how technology is used within government.

But according to nearly 20 insiders and hundreds of pages of court documents reviewed by Reuters, what’s happening inside DOGE looks less like reform and more like a shadow state.

Surveillance, Algorithms, and Loyalty

At the center of the controversy is DOGE’s alleged use of artificial intelligence to monitor internal communications across at least one federal agency. According to three people familiar with internal discussions at the EPA, DOGE operatives are rolling out tools to scan Microsoft Teams chats and emails for signs of sentiment considered hostile to Trump or Musk.

“We have been told they are looking for anti-Trump or anti-Musk language,” said one source with direct knowledge of the EPA’s internal guidance.

EPA officials later confirmed they were exploring AI to “optimize agency functions,” but denied using it for personnel decisions. Still, the climate inside the agency has shifted dramatically. Since January, more than 600 workers have been placed on leave, and the agency is preparing to cut 65% of its budget.

Ethics experts say this raises red flags about free speech and political intimidation within federal institutions. Kathleen Clark, a professor at Washington University in St. Louis, put it bluntly. “It sounds like an abuse of government power to suppress or deter speech that the president of the United States doesn’t like.”

Behind Signal and Grok

Secrecy defines much of DOGE’s operations. Staffers routinely communicate using the encrypted messaging app Signal, which allows messages to disappear. That practice could violate federal record-keeping laws, which require documentation of official communications.

And yes, this is the same app where Donald Trump’s national security adviser Mike Waltz included a journalist in a group chat about plans for US strikes in Yemen. It has now emerged that Waltz mistakenly saved the number of Jeffrey Goldberg, the editor of The Atlantic, under the contact of someone else he intended to add, according to three people briefed on the matter by The Guardian.

“If they’re using Signal and not backing up every message to federal files, then they are acting unlawfully,” Clark said.

Adding to the opacity, DOGE employees are also said to collaborate on official documents via shared Google Docs. This method bypasses traditional vetting procedures and chains of custody — an unusual workflow for government policy, where edits are typically logged and traceable.

These practices are part of a broader pattern. After Musk’s team took over the Office of Personnel Management (OPM) in January, they locked out more than 100 technical staffers from key cloud systems, including databases containing sensitive information about millions of current and former federal employees.

Today, only two people reportedly retain access to that data: one long-time federal worker, and Greg Hogan, a political appointee and former AI startup executive now serving as OPM’s chief information officer.

What is DOGE Actually Doing?

The full scope of DOGE’s technological apparatus remains murky. Insiders say the team has “heavily” deployed Musk’s own AI chatbot, Grok, as part of its cost-cutting mission. But what Grok is actually doing inside government systems is unclear — even to some of the people working with it.

Musk, who holds a special government employee status, is legally barred from using his role to benefit himself or his companies. Yet the lines are blurring. One person with direct knowledge of Musk’s earlier plans said he envisioned using government data to “build the most dynamic AI system ever.” That system, they said, could potentially “do the work” of public servants.

The Trump administration insists DOGE is exempt from public records laws because it operates under the Executive Office of the President. But a federal judge disagreed. On March 10, the court ordered DOGE to begin turning over records to the watchdog group Citizens for Responsibility and Ethics in Washington. So far, no documents have been delivered.

“An Experiment With No Oversight”

To some, the DOGE project is a long-overdue modernization effort — an attempt to drag the federal government out of its paper-clogged past and into the digital age. Few dispute that the government’s legacy computer systems are outdated and inefficient.

But critics warn that modernization is coming at the cost of transparency, legal norms, and the rights of civil servants. They say the administration is replacing career experts with loyalists, bulldozing institutional safeguards, and weaponizing new technologies to pursue ideological goals.

“This isn’t just about efficiency,” Clark said. “This is about power—who has it, who watches, and who gets watched.”

As DOGE’s influence spreads through federal agencies, the line between Silicon Valley disruption and government accountability is rapidly eroding. For now, much of what happens inside DOGE remains hidden from public view.

But the signals — both literal and metaphorical — are growing louder.

share Share

Scientists Just Built a Mini Human Nervous System That Can Process Pain in a Dish in World First

This lab-made nervous system shows how we feel pain — without hurting anyone.

This Ancient Runestone Might Be the Oldest Ever Found — and It’s Full of Mystery

Its cryptic inscriptions could rewrite the early history of runic writing in Scandinavia.

Trump-Appointed EPA Plans to Let Most Polluters Stop Reporting CO2 Emissions

One expert said it's like turning off a dying patient's monitor.

Denisovan Jaw Found in Taiwan Strait Changes the Human Migration Map

Our elusive ancient cousins once roamed much further east than previously believed

The secret to making plant-based milk tastier and healthier: bacteria

Instead of masking off flavors with sugar, salt, or artificial additives, companies can let bacteria do the work.

A 30,000-Year-Old Feather Is a First-of-Its-Kind Fossil

A new analysis of a fossil found in 1889 has unveiled the presence of zeolites—and an entirely new mineralization method.

This Sensor Box Can Detect Deadly Bird Flu in 5 Minutes. But It Won't Stop the Current Outbreak

The biosensor can detect viral airborne particles.

In 2013, dolphins in Florida starved. Now, we know why

The culprit is a very familiar one. It's us.

Researchers can't rule out the possibility of life existing on Titan

It wouldn't be very much, but it's exciting anyway.

The Earth's oceans were once green. Then, cyanobacteria and iron came in

A pale green dot?