Flock Safety AI Surveillance Tool Exposed: What Builders Need to Know

Flock Safety has long told the public its cameras "cannot recognize, identify, or track individuals." A WIRED investigation using code found on Flock's own website now shows otherwise. The reconstructed tool, OS Investigate (formerly Night Shift), can analyze movement patterns across a network of more than 6,000 communities to link license plates to names, home addresses, and family members. For AI builders, this is a clear reminder that capability often runs ahead of transparency and that ethical design must be baked in from the start.

The OS Investigate Tool: Capabilities vs. Public Claims

WIRED assembled a mock version of the tool using code from Flock's own site. The investigation found 69 suggested prompts across more than 450 files, covering queries that go far beyond plate recognition. According to the reconstruction, the system draws on police case files and data-broker information to connect plate numbers to personal details, flag possible witnesses based on repeated appearances, and suggest associates when cars are seen together repeatedly. Flock told WIRED the product is still in development and may not reflect the final version released to law enforcement, but WIRED's findings directly contradict the company's public stance.

Why This Matters for AI Builders

This case illustrates a fundamental tension: the same AI capabilities that enable powerful public-safety tools also create risks of mission creep, re-identification, and civil-liberties violations. Flock has admitted it was too slow to put safeguards in place against police misuse, a warning for any team building inference systems with access to sensitive location data. When movement patterns become investigatory leads, routine activities can be flagged without probable cause. The Institute for Justice database tracks automated license plate reader abuse, with documented incidents in 2026.

Practical Guidance for Building Surveillance AI

If you are building AI for public safety or any use case involving location-based inference, the Flock case suggests several design principles:

• Data minimization. The broader the camera network and the richer the data sources pooled, the harder it is to avoid identity inference. Limit data retention and access scope.
• Auditability. Provide clear logs of every query, who ran it, and why. Flock recently updated its recommended data storage timeline to seven days and claims to have expanded audit requirements, but external scrutiny remains essential.
• Transparent policy limits. Public claims should match actual capabilities. When a product can infer identity, that must be disclosed to oversight bodies and the communities affected.
• Guardrails against re-identification. Even if you do not intend to track individuals, movement-based profiling can effectively identify them when combined with external data. Plan for that consequence explicitly.

Limitations and Unknowns

It is important to note that the evidence comes from investigative reporting, not an official specification sheet. Flock maintains the tool is in development and the version WIRED reconstructed may differ from what is eventually deployed. Without access to the production system, builders cannot verify the exact capabilities. Still, the 69 prompts and 450 file references suggest the direction of internal development is clear. The company has also faced public backlash, including vandalized cameras and calls for stronger oversight.

For AI builders, the takeaway is straightforward: when you build systems that can tie identity to movement, you inherit the responsibility to design for accountability from day one, not after the first abuse report.