Scanned at the sliding doors: How SAFR Guard's facial recognition is quietly tracking grocery shoppers

A shopper reaches for a cart, glances at the weekly deals sign, and strolls through the entrance. In the time it takes to pocket a phone, a camera has already captured that face, built an encrypted biometric template, and compared it against a store watchlist. This is grocery shopping in 2025, and it is already happening at several Grocery Outlet locations across the San Francisco Bay Area, according to reporting from KQED and Biometric Update.

SAFR Guard launched in April 2025 as an AI-powered surveillance app built for retail loss prevention. Signs posted at participating store entrances inform customers the technology is active. That notification is, effectively, the full extent of shopper involvement in the decision.

The mechanics are straightforward. An edge-computing camera scans faces at entry and checks each one against a watchlist built from retailer-submitted incident reports covering suspected shoplifters and violent offenders. SAFR reviews each submission before anyone is added. For cases involving violence or high-dollar theft, records can be shared across participating stores within the same geographic area - a neighborhood watch program, but with biometric teeth.

Here is where SAFR’s reassurances and its own policies begin to diverge. According to Biometric Update reporting, the data retention terms break down like this:

• Non-match face templates are deleted within one second, per SAFR
• Encrypted images of non-flagged shoppers are retained for up to 30 days
• Local camera footage is stored for 30 days
• Flagged individuals’ records are retained for up to 12 months - and longer if SAFR determines continued retention is necessary

President Charisse Jacques said “The software is not a law enforcement system and has no connection with law enforcement, and it does not build a general-purpose face matching database of the public."

A Sign on the Door Isn’t Consent

Posting a notice and obtaining meaningful agreement are two very different things - and that distinction matters as retail facial recognition moves from back offices to front entrances.

The real shift is not the technology itself. Facial recognition has existed in retail back offices for years. What is new is stationing it at the entrance - scanning every face on arrival, rather than reviewing footage after an incident. That gap is the difference between a passive security camera and a system that processes your biometrics before you have touched a single piece of produce.

Passive notification - a posted sign - satisfies some legal thresholds. But as biometric privacy laws tighten across multiple states, the distance between “we informed you” and “you consented” will face growing legal scrutiny from cases involving secretly tracking users. Deployments beyond the Bay Area remain unconfirmed by available reporting.

The technology demonstrably works. Whether the rules governing retention periods, cross-store data sharing, and watchlist accuracy can keep pace with its expansion is the question nobody picking up bananas signed up to answer.