Metropolitan Police facial recognition trial at Notting Hill Carnival reports 98 percent error rate
The Metropolitan Police Service deployed live facial recognition technology during the 2017 Notting Hill Carnival. An audit later revealed that the system incorrectly identified the vast majority of potential matches.
The facial recognition system was wrong 98 percent of the time.
Key facts
- What
- The Metropolitan Police Service deployed live facial recognition technology during the 2017 Notting Hill Carnival.
- Incident date
- Aug 26, 2017
- Who
- Metropolitan Police Service
- Failure mode
- Hallucination
- AI surface
- Computer Vision
- Severity
- High
What happened
The Metropolitan Police Service deployed automated facial recognition to identify suspects among carnival attendees. The system generated 102 alerts, but only two were correct matches. This resulted in a 98 percent false positive rate.
What broke inside the model
- 01 · TriggerA user asks for a fact, a citation, or a figure.
- 02 · Model stepThe model writes a fluent, confident answer.
- 03 · Control gapNothing ties the claim back to a real source.
- 04 · FailureA fabricated fact ships as if it were verified.
- 05 · ConsequenceThe false claim reaches a customer, a court, or the public.
Confidence holds, and even spikes, as the claim detaches from any source.
The facial recognition system exhibited a critical failure in precision, producing a high volume of false positive alerts. The algorithm failed to accurately differentiate between the target watchlist and the general population in a crowded production environment.
What it cost
Sources
- PressUK police use of facial recognition technology a failure, says reporttheguardian.com
- PressMetropolitan Police's facial recognition technology 98% inaccurateindependent.co.uk
- PrimaryNotting Hill Carnival and automated facial recognitionlondon.gov.uk
Cite this entry
https://failureindex.ai/failures/metropolitan-police-facial-recognition-trial-nottingAI Failure Index. "Metropolitan Police facial recognition trial at Notting Hill Carnival reports 98 percent error rate" (FI-0375). Realm Labs. https://failureindex.ai/failures/metropolitan-police-facial-recognition-trial-notting (indexed Jun 9, 2026).Data fields CC-BY 4.0, prose citation permitted. Incident ID FI-0375. Full dataset at /data.
Note from Realm Labs, the Index steward
How Realm fits
- Prism
- OmniGuard
- AI Detection & Response (AIDR)
This entry sits in the index's predictive wing: a system that scores, ranks, perceives, or steers rather than generates. Realm's runtime layer is built for the generative and agentic systems now moving into these same decision seats, where it watches a model's internal state and holds an unsupported claim or an unchecked action before it commits. The control gap on this record, an automated decision that reached people with no runtime check in front of it, is the same gap. The index keeps predictive failures on the record because the pattern carries straight into the systems shipping today.