Flock Safety's AI "Suspicion" Alerts: What Automated License Plate Reader Escalation Means for Fourth Amendment Exposure

Automated license plate readers were sold as an investigative accelerant: an officer already focused on a vehicle could find where else it had been photographed. Flock Safety and comparable vendors have crossed a different line — systems that scan movement patterns across a national database and proactively alert agencies when an algorithm decides a vehicle looks suspicious. The shift moves ALPR from a lookup tool into a suspicion-generation engine, and that distinction matters for Fourth Amendment analysis, policy drafting, and audit design.

From Lookup to Alert: What Changed

Traditional ALPR deployment records plate reads at fixed camera locations. When investigators have probable cause tied to a specific vehicle, they query historical hits — who was near the scene, when, and in what direction.

Flock’s expanded product set, described in company materials and challenged by civil liberties advocates, adds analytics layers on top of that nationwide dataset:

• Multi-State Insights — alerts when a vehicle registers reads across state lines within a defined window, framed around trafficking and organized retail crime patterns
• Convoy and linked-vehicle searches — identifies plates frequently co-occurring in the same camera views
• Pattern-based suspicion scoring — movement histories evaluated against models that flag “suspicious activity” without an originating investigative predicate tied to that driver

The ACLU’s analysis frames the core issue: cameras that used to record comings and goings now evaluate each driver’s pattern and push alerts when the model fires. Probable cause traditionally precedes the search. Here, the search — or something functionally similar — may precede individualized suspicion.

Fourth Amendment Pressure Points

Court doctrine on ALPR remains unsettled at the federal level, but state courts are drawing lines.

The Massachusetts Supreme Judicial Court’s 2020 decision in Commonwealth v. McCarthy held that brief, targeted ALPR lookups did not always require a warrant, but signaled that sustained or dragnet collection could cross into Fourth Amendment territory when location histories reveal private details comparable to GPS tracking.

When a vendor’s central database aggregates reads from thousands of jurisdictions and runs AI pattern detection across all of them, agencies inherit exposure to arguments that:

1. The database itself is a general warrant — a pre-built index of movement without individualized authorization
2. Algorithmic alerts are searches — triggering investigative contact based on vendor analytics, not officer observation
3. Retention and sharing expand the search — a plate read captured for local parking enforcement becomes input to a multi-state suspicion model

Agencies cannot assume the vendor’s terms of service satisfy constitutional minimums. Contract language does not bind courts.

Policy Guardrails Agencies Should Document Now

Whether or not your jurisdiction deploys Flock specifically, the suspicion-alert model is migrating across the ALPR industry. Policy should address it explicitly rather than inheriting vendor defaults.

Require human authorization before contact. An algorithm flag should not, by itself, authorize a stop, knock-and-talk, or extended surveillance. Document who reviews alerts, what corroboration is required, and how false positives are logged.

Set retention and purpose limits. Reads collected for auto-theft recovery should not silently feed cross-state pattern analytics unless council or commission action authorizes that secondary use. Our ALPR retention comparison guide maps how states are diverging on deletion schedules.

Publish audit trails. Flock markets audit logs; agencies must actually review them. Rhode Island ACLU materials reference cases where audit trails exposed misuse — chiefs querying ex-partners’ plates, for example. Quarterly audit review belongs in policy, not in a slide deck vendors show once at procurement.

Ban fishing expeditions. Prohibit queries keyed only to neighborhood, time window, or demographic proxy. Tie every search to a case number or documented exigent circumstance.

Track disparate impact. Pattern models trained on enforcement-heavy corridors will flag communities that already experience over-policing. Document demographic review of alert outcomes annually.

Vendor-Neutral Procurement Questions

RFP language should survive vendor rotation:

• Does the platform generate proactive alerts not tied to a user-entered plate or case ID?
• Where is data stored, who can query it, and which other agencies contribute reads?
• Can the agency opt out of national analytics while keeping local camera utility?
• What is the false-positive rate on suspicion alerts, measured independently?

Compare answers against your state’s ALPR statute — Virginia, New Hampshire, and California impose distinct constraints that federal marketing decks rarely mention.

What Investigators Should Tell Command Staff

Detectives will want the new tools. The sales demo is compelling — trafficking cases solved, plates swapped across state lines caught in hours. None of that eliminates the constitutional and community-trust cost of alerting on movement patterns before a crime is reported.

Command staff should treat suspicion alerts like facial recognition re-identification or predictive policing heat maps: capability with a documented approval threshold, not a default background process. Legal counsel should review alert workflows before deployment, not after the first lawsuit.

The technology is not going back in the box. Agencies that write the guardrails now — authorization, retention, audit, impact review — retain investigative benefit while limiting the exposure that comes from letting a vendor’s algorithm decide who looks suspicious on your streets.