Source: Ars Technica
This case exposes how law enforcement agencies treat algorithmic confidence scores—93% in this instance—as near-dispositive evidence rather than investigative leads, then compound the error by withholding evidence that could have exonerated the defendant. The suit tests whether courts will hold police accountable for the gap between how facial recognition actually works (probabilistic, error-prone, biased against darker skin) and how departments deploy it (as near-certainty grounds for arrest). A win could force departments to document their matching methodology and the human investigation that should follow, rather than letting the tool's imprimatur substitute for evidence.