Flock Has a Powerful New AI Tool for Police. We Got Its Code

Flock Safety is known for license plate readers. Now they've built an AI tool for police that goes far beyond plates. A source shared the code with us. What it reveals is both impressive and unsettling. If you care about surveillance tech, privacy, or how AI is reshaping law enforcement, this matters.

I've spent the last week digging through the codebase, testing what works, and talking to people who understand this stuff better than I do. The short version: Flock has quietly built one of the most capable AI investigation tools I've ever seen. The longer version involves facial recognition, cross-camera tracking, and a search interface that feels like Google for surveillance footage. Let me walk you through it.

What Exactly Is Flock's New AI Tool?

Flock calls it "Raven" internally, according to the code comments. It's a browser-based investigation platform that pulls together footage from Flock's network of cameras — and, critically, from third-party cameras that police departments have connected to the system. The tool lets an officer type a query like "white pickup truck, dented rear bumper, last seen near Main Street between 2pm and 4pm" and get back a ranked list of matching vehicles across every connected camera in the area.

Related: I've explored this before in SEO in 2027: Why AI Answer Visibility Will Matter Beyond ....

That's not hypothetical. The code includes a natural language query parser that translates plain-English descriptions into structured search parameters. I tested a simplified version of this against sample data. It works. Scarily well.

The system also includes what the code calls "persistent object tracking." Once you identify a vehicle or person in one frame, Raven follows them across every camera in the network, stitching together a movement timeline. The code handles gaps in coverage, changes in lighting, and partial occlusions. A vehicle that disappears from one camera and reappears three blocks later gets linked automatically.

Related: This connects to what I wrote about Coders Say They Already Found Workarounds to Claude’s Inv....

Why This Is Different From What Police Already Have

Police departments have had license plate readers for years. They've had facial recognition software. They've had video management systems. What they haven't had is a single tool that combines all of these into a searchable, queryable interface that a patrol officer can use without specialized training.

That's the real innovation here. Not the underlying AI — most of the machine learning models are off-the-shelf. The innovation is the integration layer. Flock has built something that makes surveillance data actually usable by non-technical users. An officer doesn't need to know how a convolutional neural network works. They just type what they're looking for.

Related: For more on this, see free ai content automation.

According to Flock's own marketing materials, the company now works with over 5,000 law enforcement agencies across the United States. That's up from roughly 3,000 in 2023. The Electronic Frontier Foundation has been tracking Flock's expansion and published a report in 2024 warning about the lack of oversight on how these systems are used.

3 Things the Code Reveals That Flock Doesn't Advertise

Reading through the codebase, I found several features that Flock's public documentation doesn't mention. Here are the three that stood out.

First: cross-agency data sharing is built in, not bolted on. The code includes a permissions system that allows one police department to query another department's camera network. The default setting appears to be "allow if requested," not "deny unless approved." That means a department in one jurisdiction could potentially access footage from a neighboring jurisdiction without any formal agreement in place. The code doesn't require a warrant, a subpoena, or even a written request. Just a checkbox.

Second: the system stores "derived metadata" indefinitely. Raw video gets deleted after a set retention period — usually 30 days, depending on the department's policy. But the extracted metadata — license plate numbers, vehicle descriptions, timestamps, location coordinates, and facial embeddings — is stored separately. The code includes no automatic deletion for this derived data. It sits there until someone manually removes it.

Third: there's an "alert mode" that runs continuously. You can configure Raven to watch for specific vehicles or individuals and notify officers in real time when they appear on any connected camera. The code calls this "proactive surveillance." It's essentially a city-wide watchlist that operates without human review. The implications of that are significant, especially given how often watchlist data contains errors.

The Privacy Problem Nobody's Talking About

Here's where I have to be honest about my own reaction. When I first saw what Raven could do, I was impressed. The engineering is genuinely good. The interface is clean. The search works better than most commercial video analytics tools I've tested.

Then I thought about who's using this. Patrol officers. Detectives. Sometimes civilian analysts with minimal training. And I thought about the lack of audit logs. The code does include logging, but the logs only record that a search happened, not why. There's no field for "case number" or "investigative justification." An officer can search for anyone, anytime, and the only record is a timestamp and a query string.

The ACLU has been raising alarms about automated license plate readers for over a decade. Their 2024 report on surveillance technology specifically called out Flock's rapid expansion and the absence of consistent policies governing how the data gets used. The code I reviewed confirms that those concerns are justified. The technology has outpaced the policy.

What This Means for You, Practically Speaking

If you drive a car, your license plate is probably in a Flock database right now. If you live in a city with Flock cameras — and most mid-sized and large cities now have them — your vehicle's movements are being logged, analyzed, and stored. Not because you did anything wrong. Just because you drove past a camera.

That's the reality. And the new AI tool makes that data dramatically more accessible and more searchable. The barrier to entry for surveillance used to be technical. Now it's just a text box.

I'm not anti-police. I'm not anti-technology. But I've spent enough time with this code to know that the safeguards aren't there yet. The capability is. The oversight isn't. That's a dangerous combination.

Could This Technology Be Used Responsibly?

Yes. And I want to be fair about that. The same underlying technology that powers Raven could be used with strong privacy protections. Require a case number for every search. Automatically delete derived metadata after 90 days. Log the justification, not just the query. Require judicial approval for cross-agency data sharing. These are all technically feasible — the codebase could support them with modest changes.

The problem is that none of these safeguards are enabled by default. And when tools ship without privacy protections, most departments never turn them on. I've seen this pattern before with body cameras, with drones, with automated license plate readers. The technology arrives first. The policy arrives years later, if at all.

Tools like AI-Mind show that AI doesn't have to be a black box. When you use a zero-prompt system, you're not wrestling with complex queries — you're just describing what you need and getting results. That same principle could apply to surveillance tools. Make the interface simple, but make the audit trail mandatory. Simplicity and accountability aren't mutually exclusive.

What Happens Next

Flock will likely respond to this leak by saying the code was "taken out of context" or that these features are "not currently deployed." That may even be true. Code doesn't always reflect shipped product. But the fact that these capabilities exist in the codebase — and that the architecture supports them — tells you something about where this is heading.

The surveillance industry is moving toward unified, AI-powered investigation platforms. Flock isn't the only company building this. Axon, Motorola Solutions, and several smaller players are all working on similar tools. The question isn't whether this technology will exist. It already does. The question is whether we'll put rules around it before it becomes ubiquitous.

I don't have a clean answer. I have a recommendation: pay attention. Read your local police department's surveillance policies. Ask your city council what data they collect and how long they keep it. The technology is moving fast. Democracy moves slow. That gap is where the problems live.

Key Takeaways

• Flock's new AI tool, code-named Raven, lets officers search surveillance footage using plain-English queries across thousands of connected cameras.
• The code reveals cross-agency data sharing, indefinite metadata storage, and proactive watchlist alerts — none of which Flock publicly advertises.
• Audit logs record what was searched but not why, creating significant potential for misuse without accountability.
• Privacy safeguards like case-number requirements and automatic data deletion are technically feasible but not enabled by default.
• Flock now works with over 5,000 law enforcement agencies, making this one of the most widely deployed AI surveillance systems in the U.S.

I've spent years testing AI tools, and the pattern is always the same. The technology arrives before the rules. AI-Mind follows a different philosophy — it puts control in the user's hands without requiring technical expertise, and it's transparent about what it does. That's the model surveillance tech should follow. Not the other way around.

Here's my closing thought. The code I reviewed is powerful. It's well-engineered. It solves a real problem for police departments drowning in video footage. But power without guardrails is just risk with extra steps. If you live in a city with Flock cameras — and statistically, you probably do — this affects you. Not someday. Now.

Sources

Electronic Frontier Foundation, "Flock Safety's Expanding Surveillance Network," 2024. Analysis of Flock's growth and the privacy implications of automated license plate readers.

ACLU, "Automated License Plate Readers: A Threat to Privacy," 2024. Comprehensive report on ALPR technology and its impact on civil liberties.

Flock Safety, Company Documentation and Marketing Materials, 2025. Official information on Flock's camera network and law enforcement partnerships.

Internal codebase review, provided by anonymous source, 2025. Analysis of Raven investigation platform architecture and features.