What Actually Happens to Your Flock Data After 30 Days?

On August 10, 2026, a week before the county commissioners meet, Sheriff Ed McMahon announced that the New Hanover County Sheriff’s Office had cut its Flock data retention from 90 days to 30. It is a real change, and we’ll credit it plainly. It also lands as reassurance — the impression most people take away is that after 30 days, the data about them is simply gone. That impression is partly true and importantly incomplete. Here is what actually happens, using only Flock’s own published words, the county’s own signed contract, and independent reporting — none of which requires the Sheriff to answer a single question.

What the 30-day deletion really covers — and does well

Start by giving Flock its due. Flock’s own deletion guide is specific: license-plate data is stored in Amazon’s cloud and automatically deleted by an S3 “lifecycle policy” once it passes the retention window — 30 days by default here. Deletion isn’t a person who might forget; it’s an automated process, logged, and tested annually by outside auditors (SOC 2 Type II). Flock says that once an object is deleted, “it cannot be recovered — not by the customer, not by Flock, not by AWS.” For the raw plate captures — the images and reads themselves — that is a genuine, verifiable safeguard, and we won’t pretend otherwise.

But read the object of that sentence carefully. The 30-day clock runs on the plate captures. It is not the whole story of what the system holds about your movements.

What it does not touch — by Flock’s own account and the county’s own contract

1. The record of who searched persists. The Network Audit — the log of who ran a search, when, for what stated reason, and how far it reached — is a different record with a much longer life. We know this for a fact in New Hanover, because the county produced 16.5 months of it (2,980,082 rows, back to January 2025) at a time when plate retention was 90 days. Flock’s own deletion guide even markets this persistence as a feature, inviting buyers to ask, “Can I see exactly who has searched my data, when, and for what stated purpose?” The 30-day number never applied to the audit trail. The record we are fighting to see — who has been searching — is precisely the record that persists.

2. Flock’s own list of exceptions. In the same guide, Flock tells buyers to “know the exceptions”: data “may be retained longer when required by law, subject to a legal hold, or when a customer has a contractual or regulatory obligation to preserve specific records.” So even Flock frames 30 days as the default, not an absolute.

3. Exported copies are not recalled — and the county agreed to that in writing. When Flock data is shared to an outside system, deletion at the source does not reach the copy. New Hanover’s own contract says so: § 4.4 of Contract #25-0364 provides that a receiving third party may keep the data “for a period longer than Flock’s standard retention period,” and the county “hereby provides consent to such retention period.” The county has already agreed that copies of its data can outlive the 30-day window in other hands. (How the data leaves in the first place is its own story.)

4. Flock keeps an anonymized derivative — perpetually. Section 4.3 grants Flock a “worldwide, perpetual, royalty-free right to use and distribute” anonymized, aggregated data derived from the captures — expressly including “the training of machine learning algorithms.” Flock’s own buyer checklist gives the game away here: it tells agencies to ask whether “the contract prohibit[s] the vendor from using customer data to train models without explicit, written consent.” New Hanover’s contract contains no such prohibition — it affirmatively grants that right, and forever. So while the raw captures roll off in 30 days, a processed derivative of this county’s vehicle data persists with Flock indefinitely.

“Anonymized” is not the same as “gone”

The word doing the heaviest lifting in all of this is anonymized. It sounds harmless. Independent reporting suggests it is less so. 404 Media reported, from leaked material, that Flock has been building a people-lookup tool — pulling in outside and breached data to connect vehicles to identities. And Flock’s own product, the “vehicle fingerprint,” already identifies a car by make, color, body type, and features like a roof rack or bumper sticker — no plate required. Data that is “anonymized” in a database can still be re-attached to a person when the tools are built to do exactly that. Retention limits on the raw plate reads do not govern a derivative designed to survive them.

And the data does not stay in New Hanover

Finally, deletion at one agency does nothing about copies that have already spread. The EFF has documented that ALPR data routinely flows to hundreds of other agencies — often 800 or more — and that Flock’s national lookup network lets far-flung departments query local reads without a warrant. 404 Media has reported federal access, including ICE, reaching the network through local police. Once a New Hanover read has been queried or shared out across that network, the county’s 30-day clock cannot reach the copies it produced elsewhere. (This is exactly why the redacted searching-agency field matters: it is the record of that spread.)

What we are not claiming

We hold ourselves to the standard we ask of the county, so two limits. First, we are not claiming the county secretly keeps the plate reads past 30 days, or that the “vehicle fingerprint” is a hidden object that survives the deletion. Flock says the LPR record — including those vehicle attributes — is deleted with the capture, and we take that at face value for the origin copy. Our point is narrower and fully documented: the retention limit governs one copy of one record type, while the search logs, the exported copies, and the anonymized derivative are governed by different rules — longer ones. Second, we credit the 30-day deletion as a real improvement over 90; less retention of raw captures is better, not worse. The problem is the gap between the reassurance and the architecture, not the retention cut itself.

The honest bottom line

“Deleted after 30 days” is true of the raw plate captures, and Flock can prove it for that copy. It is not true of the audit of who has been searching (which persists, and which the county produced going back well over a year), of copies exported to other systems (which the county’s own contract lets them keep longer), or of the anonymized derivative Flock is entitled to keep forever. A resident hearing “we cut it to 30 days” reasonably concludes the record of their movements is gone. The record shows that is the beginning of the answer, not the end of it.

The public can speak at upcoming meetings. “How long do you keep the plate reads?” has a comforting answer. The better questions are: How long do the search logs live? Which outside systems hold exported copies, and under what retention? And what, exactly, is the anonymized derivative Flock keeps — and can it be re-identified? (How to ask them.)

Sources

• Flock Safety — “How Does Flock Handle License Plate Data Deletion?” (the 30-day AWS lifecycle deletion, the exceptions, the “train models” buyer question, and “the customer owns the data”) · Trust Center — Data Privacy · LPR Policy.
• New Hanover County Contract #25-0364, §§ 4.1–4.4 (Customer Data owned solely by the county; third-party retention consent; perpetual anonymized-data / ML-training right), in the archived production.
• EFF — “Flock Safety’s Feature Updates Cannot Make ALPRs Safe” · ALPR street-level surveillance (data shared with hundreds of agencies).
• 404 Media — Flock building a people-lookup tool · ICE taps the nationwide network. ACLU — Flock’s aggressive expansions.
• DeFlockILM — the 16.5-month audit the county produced · how data leaves Flock · the redacted searching-agency field.

Retention behavior described here is drawn from Flock Safety’s published materials and New Hanover County’s signed contract as of August 2026; configurations can change. We credit the county’s reduction of raw-capture retention from 90 to 30 days as a real improvement. General information and opinion for public discussion, not legal advice. Published August 12, 2026.