85% accurate is doing a lot of hiding LOL. Searching a multi-million-face gallery and even high per-comparison accuray turns into mostly false positive. THese systems are only ever defensible as an investigative lead, neve as probable cause.
That's why we don't trust them alone or can demand tests from different sources. AI, however, gets sold as an ultimate cure. Just like anything computers touch, it is assumed infallible.
This isn't just AI misidentification. This is also an eye witness picking him out of a lineup. This is really AI extending the reach of the already sketchy eye witness practice.
This is exactly like that case from Fargo earlier this year. We got a new police chief after this, but she still hasn't been compensated and nobody got in trouble for it.
This will happen more often in many domains, and it raises the general question of liability.
Should it be the AI company that created the model? The company that build the face recognition software using the model? The police department that decided to use the face recognition software?
I would assume the police department is the one legally liable, though they may turn around and sue the software company, and I guess the question is whether they can sue the frontier model company.
https://www.mprnews.org/story/2026/03/18/fargo-polices-use-o...
[0] https://www.imdb.com/title/tt31050594/
This will happen more often in many domains, and it raises the general question of liability.
Should it be the AI company that created the model? The company that build the face recognition software using the model? The police department that decided to use the face recognition software?
I would assume the police department is the one legally liable, though they may turn around and sue the software company, and I guess the question is whether they can sue the frontier model company.
These clowns need to be taken for all the money they can