BTP LFR Trial Yields One False Alert

This looks less like an accuracy win than an evaluation failure: one false alert is still consequential, while zero verified matches makes the system’s operational value impossible to establish.

• –NEC’s NeoFace M40 compares live feeds against police watchlists, with officers—not the model—making intervention decisions.
• –The £320,786 cost and nearly 100 police hours produced no attributable arrests, though results depend heavily on watchlist quality, location, timing, and deployment strategy.
• –A low false-alert count does not measure missed matches or broader demographic performance; BTP needs transparent sensitivity, subgroup, and false-negative reporting.
• –The ICO has warned that police facial-recognition use still shows inconsistent data-protection governance and requires stronger oversight, accountability, and bias controls.