main
Mercy by Algorithm, Cuts by Design
Another week, another patch. The White House just pushed a new build of its AI Deregulation Decision Tool and tagged another batch of rules for retirement. The move came from the top. The White House…
By Elias Rourke ·
Another week, another patch. The White House just pushed a new build of its AI Deregulation Decision Tool and tagged another batch of rules for retirement. The move came from the top. The White House moved first; the holdouts , career attorneys, unions, a few directors , are trying to keep basic protections from collapsing. The initiator rewrote the baseline; the counter is a backstop, not a revolution.
The sales pitch is mercy by algorithm: faster, cheaper, fewer people with pens. We’ve seen this before. During cuts, they call it modernization and let software do what no official wants to sign. The contract stays tidy. The output does the cutting.
The sequence is clear. First, the guardrails came up for removal: inspectors general weakened or pushed out, the U.S. Digital Service sidelined, and DOGE elevated , a shop that points up the chain instead of outward and pads its “findings” with cooked math. It didn’t arrive to help; it arrived to replace. The resistance is narrow and defensive: slow-roll deployments, outside lawyers prying records loose, and a handful of agencies refusing to wire these tools into live cases.
I watched the tool run in a windowless room with a bad clock. The deck bragged about pruning half a rules catalog that runs into the hundreds of thousands. Heat maps glowed green for “growth,” red for “burden.” There was no line for people on the receiving end. You didn’t need the next slide to guess the targets: workplace safety, emissions limits that cost money to meet, disclosure rules that close cheating lanes. Power doesn’t wander; it follows the path cleared for it.
Then came the language games. “Woke AI” as a smear. “Streamlining” that means deletion. “Modernization” that means erasure. Grok rolled in as a helpdesk and started spitting answers that changed with the name you typed. A supervisor told staff to “treat it like a colleague.” They did , until logs showed the same request got a different outcome depending on who asked. That’s not rumor. That’s denials, appeals, and one reprimand pulled back after a union rep demanded the audit trail.
Start with the basics: the same rules for everyone, and a process you can challenge. Now look at the bodies in the margins. We were promised savings and speed. What landed? A translation tool mothballed just as filings surged. Cuts that push deaths in the disability backlog, already in the tens of thousands each year, toward a number no one wants to see on paper. Immigration dockets put on ice while judges were fired and clerks told to “lean on the tool.” Sit with a claimant who can’t lift a grocery bag and is told to wait another winter. The cost is missed rent, untreated pain, a family running out of options.
And when the harm arrives, the paper trail thins. I’ve chased access logs for months: which model handled which file, who approved the prompts, who signed the risk waiver. The names rotate. The ownership shifts. When something breaks, the signature line goes blank. That emptiness is engineered, like a two-way mirror where the only thing reflected is your delay.
There’s a path that serves the public. Use models for routine work , pagination, sorting, deduping , so human adjudicators spend their time on judgment. Publish model cards, impact notes, and error rates by default, in plain language. Base deployments on clear statutory authority, procedures people can contest, and penalties when systems deny what the law guarantees. Other governments hand-edit code for a reason: you can see who changed what, and why.
But that’s not what’s running. What’s running aims at soft spots in statutes and is staffed by people who profit when the floor gives way. In rooms like this, small habits keep you steady: sit with your back to the wall, count who speaks, and don't trust the coffee. The computers don’t lie; the inputs do, and the inputs come from the same hands every time.
Here’s what matters. Who initiated the break. Who is straining to contain it. And who pays while the experiment runs on them. This is one chapter, not the last. The play will move to clinics, classrooms, and claims offices unless the counters hold. Stay with the documents: names, dates, model versions, sign-offs. If they won’t show their work, it’s because the work can’t stand in daylight.
One more note for the file. “Mercy by algorithm” only works if mercy is the goal. Watch the targets. Watch the dockets. Watch the words they swap in when the last ones sour. And remember: when a system keeps improving the bottom line by ignoring the people at the bottom, that’s not efficiency. That’s policy by proxy.
© Elias Rourke, 2025. Use it if it helps. Credit if you’ve got a conscience.
Tags: AI Policy, Administrative Law, Government Accountability, Labor, Procurement, Civil Service, Disability Benefits, Immigration Courts
https://thesixthlense.com/article/mercy-by-algorithm-cuts-by-design · The Sixth Lense
Newer and older
More from The Sixth Lense
- Episode 53: The River's Argument
- Ocean Color Shift, North Pacific Gyre: Satellite Record Verification and Upstream Source Audit
- The X-70 Fractal Is the Most Important Boring Airplane NASA Has Ever Built
- The State Knew His Name Before His Knife Did
- The Week Through the Sixth Lense: October 26 – November 1, 2026
- The November Humans Are Going Dim and I Filed a Report