
Biometric time clocks and the law - GDPR, BIPA, CCPA, and what 'on-device only' actually means
- FaceClock Team
- Privacy
- April 22, 2026
Table of Contents
This article exists because I keep getting the same email.
It usually starts: “We were thinking of using a face recognition time clock, but our HR person flagged that we’d have to comply with [law]. Is that going to be a nightmare?”
The short answer: it depends mostly on whether the biometric data leaves the device. The longer answer involves five laws, three court cases, one 2024 amendment, and a distinction that most vendors stay vague about. So let’s go through it.
None of this is legal advice. If you have a real deployment with real risk, talk to a real lawyer. What I want is to give you enough to know which questions to ask them.
The short version
| Law | Where | What it asks of an employer using a face clock | Who can sue |
|---|---|---|---|
| GDPR, Article 9 | EU / EEA | Lawful basis (rarely consent), DPIA, minimisation, notice, deletion on request | Regulators; individuals for damages |
| BIPA | Illinois | Written notice + written release before the first scan, public retention policy, no disclosure without consent | Anyone affected - private right of action, $1,000 / $5,000 per violation |
| CCPA / CPRA | California | Notice at collection, right to know, delete, limit use of sensitive data | Mostly the regulator (CPPA) and Attorney General |
| CUBI | Texas | Notice + consent before capture, destroy within a year of purpose ending, no sale | Attorney General only, up to $25,000 per violation |
| Colorado Privacy Act (HB24-1130) | Colorado | Biometric policy, consent rules that now reach employers of any size | Attorney General |
If you only read one row, read BIPA. It’s the one with class actions attached.
What counts as biometric data
The three big regimes define it in similar but not identical ways.
GDPR puts biometric data in the special categories of Article 9: “personal data resulting from specific technical processing relating to the physical, physiological or behavioural characteristics of a natural person, which allow or confirm the unique identification of that natural person.”
BIPA (the Illinois Biometric Information Privacy Act, 740 ILCS 14) defines a “biometric identifier” as a “retina or iris scan, fingerprint, voiceprint, or scan of hand or face geometry.” It separately covers “biometric information” - anything based on those identifiers and used to identify someone. A face embedding falls in there.
CCPA/CPRA treats biometric information broadly, including “imagery of the face” from which an identifier template such as a faceprint can be extracted.
Three things follow, and they apply to every face recognition time clock on the market:
- A photo of a face, used to identify someone, is biometric data under all three.
- The embedding - the list of numbers that face-recognition software computes from an image and compares - is biometric data under all three. In FaceClock’s case that’s 128 floating-point numbers per person.
- Deleting the photos and keeping the embedding still counts as keeping biometric data.
What the laws require
These are the operational cores. There are exceptions and carve-outs; the point is the shape.
GDPR
- A lawful basis. In employment this is rarely consent, because the power imbalance makes “freely given” hard to argue. It’s more often a national law that authorises biometric time-and-attendance, or a carefully argued legitimate interest. Some member states are strict here. The Dutch regulator, for instance, fined a company €725,000 in 2020 for fingerprint-based attendance.
- A Data Protection Impact Assessment (DPIA). For systematic biometric processing of employees, assume you need one.
- Data minimisation, security measures, clear notice, and the usual data-subject rights (access, deletion).
BIPA
- Written notice and a written release before the first scan. Since the 2024 amendment (SB 2979, signed August 2, 2024), an electronic signature counts.
- A public, written retention and destruction schedule.
- Destruction when the purpose is satisfied or within 3 years of the person’s last interaction with you, whichever comes first. For a time clock, that usually means: within three years of the employee leaving, and sooner if your written policy says so.
- No selling, leasing or trading. No disclosure without consent - that’s Section 15(d).
- A reasonable standard of care in storing and transmitting.
The penalties are why BIPA gets all the attention: $1,000 per negligent violation, $5,000 per intentional or reckless one. Two cases turned that into real money. Rosenbach v. Six Flags (2019) held that a person doesn’t have to show actual harm to sue. Cothron v. White Castle (2023) held that every scan was a separate violation, which put White Castle’s theoretical exposure in the billions.
The 2024 amendment reined that in: repeated collection from the same person with the same method now counts as one violation, not one per clock-in. That lowers the ceiling a lot. It doesn’t change the basic rule. If you scan an Illinois employee’s face without a signed release first, you’re exposed.
CCPA / CPRA
- Notice at collection: what you collect and why.
- Right to know, right to delete, right to opt out of sale or sharing.
- Biometric data used for identification is “sensitive personal information”, which adds a right to limit its use.
- Since January 2023, employee data is fully in scope. Before that, it was partially exempt.
Texas and Colorado
Texas CUBI (Business and Commerce Code §503.001) is BIPA-shaped without the private lawsuits: notice and consent before capture, no sale, destruction within a year after the purpose ends. Only the Attorney General can enforce it, but it can seek up to $25,000 per violation, and in 2024 it settled with Meta for $1.4 billion under this law. It’s not a sleeping statute.
Colorado’s HB24-1130 amended the Colorado Privacy Act for biometrics, effective July 1, 2025. The part that matters here: its employer rules apply regardless of the usual size thresholds, so a 12-person Denver bakery is covered too.
How “on-device only” changes the picture
Here’s the distinction that matters most, and that marketing pages blur.
Cloud processing. The device takes a photo and sends it (or an embedding) to a server. The server matches it and sends a name back. Most “easy to deploy” face clocks work this way. The biometric data is in transit and at rest in the vendor’s cloud, so the vendor is at least a processor, and every one of the laws above now has a second party to worry about.
On-device processing. The phone or tablet captures the image, computes the embedding with a model bundled in the app, compares it with embeddings stored on that same device, and produces a match or a non-match. Nothing leaves the device. There is no server.
The laws treat these very differently:
- GDPR: no processor agreement with the vendor, no international transfer, no breach-at-vendor scenario. The DPIA and the lawful basis are still yours, but the risk section of the DPIA gets much shorter.
- BIPA: Section 15(d) is about disclosure. If the data never leaves the device, there is no disclosure to anyone. The consent and retention duties stay; the transmission exposure goes away.
- CCPA: the “sale” and “sharing” questions mostly disappear, because there’s nothing to sell or share. Right to know and right to delete become things the person holding the tablet can answer in two minutes.
It’s not magic. It moves the regulatory weight onto you, the employer, and removes the parts people were most scared of: third-party access, cross-border transfers, and a vendor breach exposing your staff.
What FaceClock actually stores, and where
Vague privacy claims are cheap, so here is the full inventory for FaceClock, taken from the app itself rather than from a brochure.
| Data | Stored where | How long |
|---|---|---|
| One reference photo per employee | App storage on the device | Until you delete the employee |
| Face embedding (128 numbers) per employee | Local SQLite database on the device | Until you delete the employee |
| Employee name | Local database | Until you delete the employee |
| Clock-in photo and clock-out photo per shift | App storage on the device | Automatically deleted after 90 days |
| Shift timestamps and duration | Local database | Automatically deleted after 90 days |
A background job runs every hour and removes shift records and their photos older than 90 days. The 90 days is fixed in the app, not a setting someone can quietly change.
Face detection uses YuNet and matching uses SFace, two open models from the OpenCV Zoo, run on the device through ONNX Runtime. The app’s Android manifest declares one permission, CAMERA. There is no INTERNET permission, so the operating system itself won’t let the app open a network connection. No analytics SDK, no crash reporter.
What “on-device” has to mean to count
Some things I see mislabelled:
“Edge processing” isn’t always on-device. Some products compute the embedding in your office and then send it to the cloud for matching. That’s still a biometric data flow to a third party.
“We delete photos immediately” doesn’t help if the embedding sits in the cloud. The embedding is biometric data under every law above.
“We don’t store biometric data on our servers” doesn’t help if it’s transmitted. Transmission is its own regulated event, especially under BIPA 15(d).
For a claim of on-device-only processing to hold, all of these have to be true:
- The matching model runs on the device.
- Embeddings live only in the app’s local storage.
- Matching works with no network at all - airplane mode, indefinitely.
- No telemetry includes biometric data, hashed or otherwise.
You can check the first and third yourself. On any Android app, the permission list on the Play Store page (or in Settings, Apps) tells you whether it can reach the internet. If INTERNET isn’t there, it can’t. For apps that do have it, you’re trusting the vendor’s documentation.
What you still have to do as the employer
Even with the cleanest architecture, the business deploying the clock is the controller. You still need to:
- Decide your lawful basis. Under GDPR this is the part to argue carefully, and it varies by country. Get local advice.
- Write a one-page notice. What you collect, why, where it lives, how long, who can see it. The table above is most of it.
- Get the signature where the law asks for it. Illinois: written release before the first scan. Texas: consent before capture. EU: consent is a weak basis in employment, so pair it with another one.
- Offer an alternative. A manual sign-in sheet for anyone who objects costs nothing and takes a lot of heat out of the conversation. In some jurisdictions it’s close to expected.
- Keep your retention rules consistent. FaceClock keeps shift data for 90 days. If your country requires working-time records for longer (Germany and Spain both have multi-year rules for hours records), export the CSV every month and keep that. The export doesn’t contain faces, so your long-term record isn’t biometric.
- Delete people when they leave. Removing an employee in FaceClock removes their profile, reference photo and embedding from the device. Put “delete from the time clock” on your offboarding checklist, next to “take back the keys”.
Three things that surprise people
Consent isn’t the silver bullet it looks like. Especially in the EU. A regulator looking at an employer-employee relationship will ask whether saying no was a real option. That’s why the alternative in point 4 matters.
Device backups can bring the cloud back in. Android can back up app data to the owner’s Google account. If the device belongs to a person rather than the business, check the backup settings, or use a dedicated tablet signed into a business account with backup turned off.
Former employees linger. BIPA’s three-year limit runs from the last interaction, so someone who left in 2024 shouldn’t still be enrolled in 2027. The 90-day purge covers shifts, not the enrolment itself - that part is on your offboarding checklist.
Where it gets hard
The easy cases are the common ones: one cafe, one dental practice, one workshop. They usually end up with a one-page notice, a signature, and a paragraph in the staff handbook.
The hard cases involve several jurisdictions at once, sector rules on top (healthcare, finance), or a works council or union agreement that adds its own process. In Germany, for example, the works council has a co-determination right over any technical system that can monitor employees, and a face clock clearly qualifies.
My recommendation
Do two things, in this order.
First, make sure the architecture is genuinely on-device. It removes most of the legal complexity. Verify it - check the permissions - rather than trusting the marketing page. Our guide to choosing a facial recognition time clock has a checklist for that.
Second, spend fifteen minutes with a local employment lawyer to confirm your setup. It costs little.
If both check out, you’re mostly done. The “biometric privacy nightmare” people worry about is mostly a cloud-processing problem, not a face-recognition problem. And if you want to see how the day-to-day record-keeping works without a cloud account, tracking employee hours without the cloud covers the rest.
Try it on a phone you already own.
FaceClock runs on most Android devices from the last five years. Install it, mount it by the door, and your team registers itself on the first clock-in. Shifts start logging the same day. No account. No card.
