Bloomington’s Facial Recognition Settlement Is a Warning Shot We’d Better Heed

Another city, another settlement, another man who lost months of his life because a algorithm said he looked like a suspect. Bloomington, Indiana just paid out to Robert Williams — a Black man arrested in 2022 after a facial recognition match tied him to a theft he had nothing to do with. The city’s insurance carrier settled for an undisclosed amount. No admission of wrongdoing, naturally. Just a check and a confidentiality clause.

The Match That Wasn’t

Here’s what happened: a detective ran a grainy surveillance still through Clearview AI — the same controversial platform that scraped billions of social media photos without consent. The system returned Williams as a “high confidence” match. The detective didn’t verify the match against other evidence. Didn’t check Williams’s alibi. Didn’t even pull his driver’s license photo for a side-by-side comparison before swearing out the warrant.

Williams spent nearly a week in jail. Lost his job at a warehouse. Missed his daughter’s birthday. All because a proprietary algorithm — trained on datasets we still can’t audit — said two faces looked similar enough.

We Keep Pretending This Is a Bug

It’s not a bug. It’s the business model.

Facial recognition vendors sell certainty. Their marketing decks boast “99% accuracy” on controlled benchmarks — well-lit, front-facing, neutral expression. Real police work offers none of those conditions. CCTV footage is compressed, angled, low-res. Suspects wear hats, masks, hoodies. The gap between lab metrics and field performance isn’t a rounding error. It’s a chasm.

And the vendors know it. Their contracts quietly shift liability to the agency. Their terms of service forbid independent auditing. Their lobbyists fight every transparency bill in every statehouse.

The Human Cost Doesn’t Scale

Williams’s case made headlines because he sued and the city settled. Most don’t. Most take a plea deal because they can’t afford a lawyer, can’t risk a trial, can’t miss another shift. The algorithm’s false positive becomes a conviction statistic. The system calls it efficiency.

We’ve seen this pattern in Detroit. In New Orleans. In Paterson, New Jersey. Black men disproportionately flagged. Cases quietly dropped. Settlements signed with NDAs. The technology marches on.

What Actual Accountability Would Look Like

Ban? Moratorium? Sure. But let’s start with basics that should’ve existed years ago:

• No arrest warrant solely on a facial recognition hit. Corroborating evidence — physical, testimonial, digital — must exist independently.
• Mandatory disclosure to defense counsel: the specific system used, its training data demographics, its known error rates across race and gender, the exact probe image quality.
• Independent auditing rights. No more trade-secret shields blocking NIST or academic evaluation.
• Strict retention limits. Probe images and match logs deleted after 30 days if no charges filed.
• Personal liability for officers who skip verification steps. Qualified immunity shouldn’t cover willful blindness.

None of this is radical. It’s basic due process adapted for a technology that fails differently than fingerprinting or DNA.

The Settlement Changes Nothing

Bloomington’s payout closes one file. The detective still has his badge. Clearview still sells to hundreds of agencies. The next Robert Williams is already being scanned right now — at a traffic stop, a protest, a mall food court.

We’re not waiting for the technology to improve. We’re waiting for the law to catch up. And every settlement without structural reform is just the cost of doing business.

Source: original report