The Robots Went to the Prosecutor First

Stephen Embry, a legal tech blogger who writes for Above the Law, published a column in July 2026 with a simple observation. Criminal defense lawyers do not have many AI tools built for them. Contract review software exists for big firms. Discovery tools exist for corporate litigation. Criminal defense gets one real specialist, a company called JusticeText, and a resource list from Berkeley Law. Embry's explanation: the market is small, the clients are poor, and public defenders spend their budgets chasing politicians instead of buying software.

All true. All incomplete.

Embry is answering the wrong question. He asks why AI didn't show up for the defense. The sharper question is where it did show up. And the answer is the other table in the courtroom.

The Market Nobody Ignored

Axon Enterprise, the company that makes police body cameras and tasers, booked $2.8 billion in revenue in 2025. Its recurring revenue crossed $1.3 billion, up 35 percent year over year. Bookings for future contracts hit $14.4 billion. Axon now sells AI tools directly to prosecutors. One product drafts police reports automatically. Another, Justice Premier, helps district attorneys synthesize evidence. San Bernardino County signed a ten-year contract. Clackamas County is shopping.

This is not a company waiting for the criminal justice market to get interesting. This is a company that already won it, on the side that pays.

Meanwhile the money going to indigent defense nationally sits somewhere around $7 to 8 billion, competing against roughly $8.5 billion for prosecution and vastly more for policing and corrections. California's own numbers, released by the state's public defense office in May 2025, show counties handing district attorneys $2.2 billion against $1.2 billion for public defenders in a single fiscal year. Call it a billion-dollar gap in one state alone.

So the "small, underfunded market" Embry describes isn't stagnant. It's losing an arms race it didn't know it entered.

The Tell

JusticeText exists because two engineering students in Chicago got angry about a police killing and started building software that lets public defenders search body camera footage the way everyone else searches a Google Doc. By December 2025 the company had $4 million in annual revenue, more than 4,100 attorneys using it, and coverage in 7 of 22 statewide public defender systems. That's real. That's also a rounding error next to Axon.

Here's the mechanism worth naming. Every camera Axon sells to a police department generates footage. Every AI tool Axon sells to a prosecutor turns that footage into leverage faster. Someone still has to review it on the other side, and that someone is a public defender carrying two or three times a sustainable caseload, using a tool built by a nonprofit-adjacent startup running on $4 million a year against a company doing $2.8 billion. The technology wave didn't skip the defense bar. It armed the other side first and left the defense bar to catch up with the runoff.

Embry's column treats "vendors don't see the money" as a natural law, like weather. It isn't.

What The Fallback Actually Costs

Embry predicts that when defense lawyers want AI, they'll turn to ChatGPT or Claude instead of paying for something specialized. He's probably right about the instinct. He's wrong to treat it as a harmless workaround.

In February 2026, a federal judge in the Southern District of New York ruled in United States v. Heppner that a defendant's conversations with Claude were not privileged and could be handed to prosecutors. A public defender who runs case notes, client conversations, or strategy through a consumer AI product is not finding a free alternative. He's creating discoverable evidence, sometimes for the same prosecutor sitting across the table with the $2.8 billion vendor.

The gap in specialized tools isn't just an access-to-justice problem. It's actively pushing defenders toward tools that can hurt their own clients.

What Breaks The Pattern

None of this is fixed rule. Three things already show it can move.

New Jersey's Office of the Public Defender launched its own open source evidence tool in July 2026, built with Princeton, free to any office that wants it. California ran a Public Defense Pilot Program worth roughly $140 million across three funding cohorts. Washington's Supreme Court cut the maximum felony caseload a public defender can carry from 150 to 47, forcing the state to either hire more defenders or fund more tools, starting in 2026 with a ten year phase-in.

Every one of these is a funding decision, not a market discovery. Nobody found a secret pool of paying customers. Somebody decided the market should exist and built it anyway.

What to Watch

Axon's prosecutor and DA contract numbers in its quarterly filings. Growth there is the clock the defense side is racing against.

Whether JusticeText or a comparable tool cracks more than 7 of the 22 statewide public defender systems, or raises new funding beyond its 2022 seed round.

Appellate courts on AI privilege after Heppner. If that ruling holds and spreads, the "just use ChatGPT" option gets more dangerous, not less.

State legislatures for indigent defense technology line items, the way California and Washington already moved. That's the lever, not vendor goodwill.

The Move

If you fund public defense, the money already works when it's earmarked, not hoped for. If you build legal tech, the market Embry calls empty has a state government as a customer with a demonstrated bill it's willing to pay, you just have to sell to the side that isn't already flush.

If you're a defender, the free AI in your browser is not neutral. It has a memory, and in at least one federal case that memory belonged to the prosecution.

The tools didn't skip criminal defense. They went where the money already was, and the money was on the other side of the aisle first.