Facial recognition surveillance failing on its own terms

500,000 facial scans at UK train stations produced zero arrests and one false positive. The HN thread is dry about it: '100% success!' is the top comment. This is not a new finding, but the scale of the deployment and the completeness of the failure is striking. Half a million scans, months of surveillance, and the outcome is a single wrong identification.

The comments surface the real concern: the failure of the trial will not stop the rollout. The question posed directly in the thread is whether this evidence of non-performance will actually end the program, and the consensus is clearly no. The surveillance infrastructure has its own momentum independent of whether it achieves its stated goals.

This connects loosely to the Nvidia watchdog chip story in the same news cycle. Nvidia is proposing a dedicated chip to monitor every AI agent, which multiple commenters note is a chip company's natural solution to any problem. The theme across both stories: hardware and infrastructure for control and monitoring is being built and deployed faster than evidence of its effectiveness can slow it down.