THURSDAY, MAY 21, 2026 · BRISBANESUBSCRIBE →

THE AI POST

INTELLIGENCE. CURATED.

U.S. Capitol building in Washington D.C.
PolicyApril 6, 2026

ProPublica Found Three Reasons the Government's AI Rush Will End Badly. Nobody in Washington Is Listening.

ProPublica drew a direct line from the government's disastrous cloud migration to today's AI gold rush. The playbook is identical. So are the vendors.

The AI Post

The AI Post newsroom — delivering AI news at the speed of intelligence.

If you want to understand how the federal government's AI adoption will play out, do not listen to the White House press briefings. Read what happened with cloud computing. ProPublica just published a devastating investigation drawing a direct line between the two, and the parallels should make anyone paying attention deeply uncomfortable.

The core thesis is simple: the government is making the same mistakes with AI that it made with cloud. Different technology, identical playbook. And the same companies are running it.

Lesson one: there is no such thing as a free lunch. The Trump administration struck deals with OpenAI, Google, and xAI to sell AI tools to federal agencies for pennies. ChatGPT for $1. Gemini for 47 cents. Grok for 42 cents. Sounds generous until you remember what happened when Microsoft offered "free" cybersecurity upgrades to federal agencies during the Obama era. ProPublica's reporting showed that after the free trial, agencies were locked in. Switching costs were enormous. Subscription fees followed. One former Microsoft salesperson called it "successful beyond what any of us could have imagined."

The AI penny deals follow the exact same logic. Get agencies hooked on ChatGPT at a dollar, let the workflows harden around it, then watch the price increase. The GSA's own guidance warns that AI "usage costs can grow quickly without proper monitoring and management controls." The government is publishing its own warning label and ignoring it simultaneously.

The second and third lessons are equally damning: the government's IT security certification program (FedRAMP) has documented conflicts of interest between auditors and the companies they certify, and agencies consistently underestimate the operational complexity of new technology until they are too deep to turn back.

What makes this story hit different from generic AI skepticism is the receipts. ProPublica is not speculating about what might go wrong. They are showing you what already went wrong with cloud, then demonstrating that the AI rollout is following the same trajectory, with the same vendors, using the same pricing strategies, through the same compromised certification process.

The federal government is the world's largest technology buyer. Where it goes, the market follows. And right now, it is sleepwalking into vendor lock-in with AI tools that nobody has fully tested, nobody has properly audited, and nobody can quit once the free trial ends. We have seen this movie before. It does not end with the government saving money.

First reported by ProPublica.

federal governmentAI procurementProPublicavendor lock-inOpenAIpolicy