The AI software prosecutors use is basically rigged to hide evidence that could prove a defendant is innocent.
March 19, 2026
Original Paper
Rebuilding Brady for Machine-Mediated Discovery
SSRN · 6432478
The Takeaway
In massive digital trials, lawyers use 'active learning' tools to find relevant documents. However, these tools are designed to find clusters of similar evidence (like a prosecution's case), whereas exculpatory 'smoking guns' are usually unique and dispersed, meaning the software is structurally optimized to ignore the very evidence that could prevent a wrongful conviction.
From the abstract
Brady doctrine is widely understood as an ethics-inflected rule against prosecutorial concealment. But in prosecutions built on digital evidence, the more consequential discovery failure is increasingly architectural. Evidence can be suppressed not because a prosecutor withholds a known document, but because the State's review workflow-triage, prioritization, and stopping-predictably fails to surface favorable material inside the government-controlled universe. When review is no longer reading b