The Vendor Audit: We Graded Flock Safety's Public Disclosures Against Its Own Documents. It Got a 43.

The lowest score on the card, because the same claim appears at three different magnitudes. The homepage headline reads "20% of U.S. crime is solved with Flock." The federal facilities page closes with "15% US crime solved with Flock." The footnoted source, Flock's own 2025 Impact Census, says something narrower than both: "In jurisdictions where Flock is deployed, an estimated 20% of solved cases were assisted with Flock technology." Those are not the same statement — "20% of U.S. crime" and "20% of cleared cases in Flock-covered jurisdictions" differ by every unsolved crime and every jurisdiction without a Flock camera. The outcome figures diverge the same way: the federal page claims "2,800 Crimes solved daily," the law enforcement page claims "2,200+ Crimes solved per week," a spread of roughly nine to one, neither footnoted. The census itself deserves credit — it discloses its survey design, sample size (roughly 700 agencies across 43 states), NIBRS anchoring, and a bootstrap range of 8,000–19,600 for the missing-persons estimate. That is more methodological transparency than this sector typically offers. The failure is not the research; it is that the headline numbers on the marketing pages do not match the research they cite, or each other.

The Trust center states the limit absolutely and repeatedly: the Data Privacy page's summary line is "It captures vehicles. Not people," listing biometric data, driver information, and facial recognition data as not collected; Myths vs Facts says "Flock LPR only captures vehicles, not biometric data or faces"; What Is Flock describes cameras "deliberately avoiding the capture of unnecessary personal details about people." The video camera product page's feature list includes "People Detection Alerts" and promises "instant alerts for people and vehicles." The FreeForm page advertises the query "man in blue shirt and cowboy hat," describes "person-related video search results," and titles its own FAQ entry "Does FreeForm people search use facial recognition?" Flock's reconciliation is technically real and stated on that page: FreeForm uses "observable, non-biometric visual characteristics... such as clothing and visible accessories," not facial recognition or biometric matching. That distinction matters — a clothing description is not a persistent identifier tied to a body, and the company draws the line explicitly where a less careful vendor would blur it. But the distinction FreeForm draws is biometric versus non-biometric person search. The distinction the Trust center draws is vehicles versus people, and the Trust center never discloses that person search exists at all. A resident who reads the privacy page and stops there has been told the system does not do something it is sold as doing.

The Trust center poses the question to itself and answers in one word: "Is there a national database of my data? No. Each agency controls its own data independently." The product catalog sells, by name, the "National LPR Network — Tap into the nation's largest crime-solving LPR network. Agencies get real-time alerts and footage from partners across the country." The LPR page advertises "billions of monthly plate reads" in "a shared network." Flock's defense rests on the word automatically, and it is a legitimate architectural distinction consistently maintained across every Trust page: sharing is opt-in, governed by written agreements, and "nothing is shared unless the agency turns it on." A federated network of consenting agencies is genuinely not the same thing as a centrally-owned federal repository, and the company never claims otherwise in its detailed pages. The score reflects the mismatch between that careful architecture and the flat one-word "No" given to a resident asking whether a national database exists — a question about existence answered as though it were a question about defaults, on the page specifically built to answer hard questions plainly.

Myths vs Facts says Flock is "auto-deleting data after 30 days"; the Data Privacy page says "Deletion is automatic." The LPR camera FAQ says Flock "will offer an extended retention for up to one year," conditioned on the agency first obtaining approval from an elected official or governing body — a real accountability requirement, and to the company's credit it is disclosed rather than buried. The What Is Flock page adds that data is deleted "unless the data is required for an ongoing investigation," an exception with no stated time limit. And the governing LPR Policy discloses a category absent from every marketing page: "Flock uses a fraction of LPR images (less than one percent), which are stripped of all metadata and identifying information, solely for the purpose of improving Flock Services through machine learning." Metadata stripping is a substantive mitigation and the policy says so plainly. But images retained to train models are not deleted on a rolling 30-day basis by definition, and the four public-facing pages that tell readers deletion is automatic do not mention the carve-out. The 30-day default appears to be real; it is the word "automatic," repeated without qualification, that the documents do not support.

The public claim is unqualified: ownership "belong[s] to and [is] retained solely by our customers," with Flock holding "a limited license to access the customer data for the sole purpose of providing our services," and Myths vs Facts adding that "no secret federal agency access or 'backdoors' exist." The LPR Policy describes two internal access tiers — CJIS-certified engineers and "priveledged administrators" (the misspelling is the policy's) — and sets the disclosure standard in a single sentence: Flock may disclose LPR data to law enforcement, government officials, or third parties "if legally required to do so or if Flock has a good faith belief that such access... is reasonably necessary to comply with a legal process, enforce the agreement between Flock and the customer, or detect, prevent or otherwise address security, privacy, fraud or technical issues." Two standards joined by "or": one adjudicated by a court, one exercised unilaterally by the vendor, with no customer-notification requirement stated. This is ordinary commercial contract language found across the software industry, which is exactly why it belongs on the card — the boilerplate is the finding. The policy also carries real countervailing commitments that a weaker vendor would omit: Flock will not "sell, publish, exchange, or disclose customer's LPR data for commercial purposes," and names a specific accountable official as custodian. The gap is between "solely" on the marketing page and "or good faith belief" in the policy that governs.

In text, Flock's compliance posture is stated consistently in two places and is unremarkable in the good sense: the Data Privacy page cites SOC 2 Type II and ISO 27001 with privacy principles "aligned to ISO27701," and the LPR Policy independently corroborates "third party attestations to requirements for SOC 2 Type II and ISO 27001." Those are specific, named, and verifiable. The federal facilities page instead displays a badge row under "Aligned to Federal Standards" containing NIST, NDAA Compliant, AICPA SOC, AICPA SOC 3, and FedRAMP. SOC 3 alongside SOC 2 is not a contradiction — SOC 3 is the public summary report of the same AICPA framework — and NIST alignment is separately claimed in the LPR Policy's security section. FedRAMP is different: a distinct federal cloud authorization with a public marketplace of authorized providers, claimed in no written statement anywhere on the site. The badges are rendered as images without alternative text, so none of these claims appear in the page's machine-readable text — meaning the site's strongest compliance assertions are invisible to screen readers and to anyone reviewing vendor claims by text search. The mid-range score reflects a real, specific, documented certification program undercut by an unsubstantiated badge displayed in the one format that resists verification.

Graded separately from feature #2 on purpose, because the narrow claim and the broad claim perform very differently. The narrow claim — that Flock does not run facial recognition or biometric identification — holds up everywhere it appears, including in the places where a company inclined to shade the truth would have gone quiet. The Trust FAQ, the Data Privacy page, the LPR camera FAQ, and What Is Flock all state it consistently. Most notably, the FreeForm page states it against interest: on the very product that searches for people by description, Flock affirmatively volunteers that the feature "does not use facial recognition or biometric person recognition" and that searches "are not designed to identify a specific person." The LPR Policy's enumerated data fields corroborate the architecture — plate image, vehicle image, vehicle characteristics, plate number, state, date, time, camera location, with no biometric field listed. Points come off for the marketing framing rather than the substance: "not facial recognition" is repeatedly deployed on the Trust pages as though it answered the broader question of whether the system watches people, which the product catalog shows it does not. On the specific question of biometrics, the disclosure is accurate and unusually well corroborated.

The strongest area on the card, and the one where the public claims are backed by specifics rather than adjectives. The Law Enforcement Access page describes individual authentication with no shared logins, role-based permissions set by the local agency, a required investigative purpose attached to each query, and automatic logging reviewable by supervisors. The LPR Policy corroborates with the actual field list captured on every query — username, date, time, purpose of query, and the plate or elements used to search — plus hardware-based multi-factor authentication for privileged administrators, TOTP for customer accounts, role-based access controls, and a named official custodian. That is a checkable mechanism, not a promise, and the field list in particular is the kind of detail that makes an agency's audit either possible or impossible. Two honest deductions. First, the system logs a stated purpose but the disclosures describe no mechanism validating that the stated purpose is truthful — the Trust page's claim that accountability "doesn't depend on people reporting their own actions" is true of the logging and not of the purpose field. Second, every oversight mechanism described is internal to the customer agency; Flock's own pages say supervisors review activity and the agency applies discipline, and the company discloses no independent or public audit of whether that review happens. The controls are real. Whether anyone outside the agency ever looks at them is not addressed.