From July 7, 2026, every newly registered car, van, truck, and bus in the European Union must ship with an Advanced Driver Distraction Warning (ADDW) system — typically an infrared camera aimed at the driver's face that tracks gaze, head position, and attention. That deadline is phase two of the EU General Safety Regulation (GSR2) — the same package that added mandatory pedestrian AEB and tire monitoring; see explainx.ai's five mandatory EU car safety features guide for the full GSR2 context and Tesla FSD overlap.
The goal is road safety. The backlash — on Reddit, Cybernews, and across EU media — is about privacy: another camera pointed at your face, another sensor in a vehicle that already uploads telemetry, and unclear rules about what happens when that video leaves the car.
This guide explains what the law actually requires, where privacy breaks down in practice, and what engineers should build — including on-device distraction detection, anonymization before upload, and API patterns for fleets, insurers, and OEM debug pipelines.
TL;DR
| Question | Answer |
|---|---|
| When did ADDW become mandatory? | July 7, 2026 for all new EU registrations (phased since 2024 for new types) |
| What does it do? | Camera tracks gaze/head; warns if you look away too long |
| Does it record me for police? | Not required — regulation targets real-time warnings, not authority upload |
| Can it use facial recognition? | No — biometric identification is prohibited for ADDW |
| Must data stay in the car? | Yes in principle — closed-loop, minimal retention, no external transmission |
| Where is the real risk? | When OEMs, insurers, or debug tools export cabin video off-device |
| US timeline? | Similar requirements expected on 2027+ model year vehicles |
| What should builders do? | Detect distraction on-device; anonymize faces before any cloud upload |
What ADDW is — and what it is not
ADDW (Advanced Driver Distraction Warning) is part of the EU General Safety Regulation (Regulation (EU) 2019/2144), implemented via Delegated Regulation (EU) 2023/2590.
What the system must do
- Monitor driver gaze and head movement while driving
- Activate automatically above 20 km/h (with provisions at lower speeds)
- Warn when attention drifts — thresholds include roughly 6 seconds off-road gaze between 20–50 km/h, and 3.5 seconds above 50 km/h (per manufacturer implementation under type-approval rules)
- Use visual distraction detection — lane-keeping alone is not enough; the driver must be observed directly
What the system must not do (on paper)
Per EU technical rules and GDPR framing:
- No biometric identification of occupants (no facial recognition login, no identity matching)
- Closed-loop operation — process only what is needed for the warning function
- No unnecessary retention — don't store driver video beyond immediate function
- No external transmission of cabin video to servers or third parties for ADDW itself
The European Commission's public messaging: "Smarter cars mean safer roads." Critics' messaging: "First it's your car, then your living room."
Both can be true. ADDW as a warning light is not the same as ADDW footage in a cloud bucket.
Why drivers are angry (and not wrong to ask questions)
The Reddit thread on EU driver cameras that hit 576+ upvotes isn't really about seatbelts. It's about precedent:
- The hardware is permanent. Once every car has a driver-facing camera, software updates can change behavior — logging, upload, insurer scoring.
- Manufacturers have a privacy track record. Reporting on automotive data misuse makes "trust us, it's local" a hard sell.
- Inspection economics. Comments from the Netherlands note annual roadworthiness checks (APK) — will a working camera become part of compliance?
- False positives hurt safety. Drivers report distraction systems that brake during overtakes or nag on long trips — safety tech that becomes distraction.
- Deactivation is limited. Regulation allows manual deactivation of warnings or the system in some configurations — but obstructing the camera can trigger other vehicle behaviors.
The law says no authority mandate to record. The fear is capability without durable enforcement — the same gap GDPR was supposed to close, now inside the cabin.
The privacy gap nobody puts in the press release
Here's the distinction that matters for engineers:
ADDW does not require storing video. The privacy risk appears when someone decides to retain, upload, or share footage.
Real-time distraction detection only needs:
- Frame-in → pose/gaze estimate → warning out
- Optionally: event metadata (timestamp, distraction score, speed) — not a recognizable face
Problems start when video is used for:
| Use case | Privacy risk |
|---|---|
| OEM debug after crashes | Identifiable driver faces in support tickets |
| Insurance telematics | Gaze scoring tied to identity or policy |
| Fleet management dashboards | Supervisors reviewing cabin clips |
| ADAS training data collection | Biometric video in ML pipelines |
| Warranty / legal discovery | Unblurred exports in evidence chains |
GDPR still applies. Face and gaze data are personal data under the same framework covered in explainx.ai's EU AI Act vs US policy guide and Europe AI landscape 2026. Minimization, purpose limitation, and deletion aren't optional because the camera is labeled "safety."
Privacy-by-design means:
- Process on-device — inference at the edge, in the ECU
- Default to no recording — metadata-only event logs
- Anonymize before any export — if video must leave the car, blur or anonymize faces first
- Separate ADDW from infotainment cloud — different trust boundary, different retention policy
Related: Flock cameras and civil liberties — same principle for footage you share; Immich and face blur before publish; license plate blur for dashcam pipelines.
US is next — same camera, same questions
Reporting from Cybernews and automotive trade press: similar driver-monitoring requirements are expected in the US on 2027+ model year vehicles, aligned with NHTSA distraction-mitigation direction.
If you're building for one market, you're probably building for both. Design for EU closed-loop rules and you are well positioned for US state privacy laws (CPRA, etc.) and insurer scrutiny.
What builders should implement
1. On-device distraction detection (no cloud required)
ADDW-class logic can run entirely in the vehicle:
- Head pose + eye gaze estimation
- Temporal smoothing (avoid single-frame false alarms)
- Speed-aware thresholds
- Haptic / audio / cluster warnings
You do not need to send raw video to a server to detect "driver looked at phone for 4 seconds."
2. Video Intelligence API — detect distraction, preserve privacy
For teams building fleet tools, aftermarket dash systems, insurer pilots, or OEM validation rigs — not every pipeline runs inside a homologated ECU on day one.
BGBlur Video Intelligence is designed for this layer:
- Upload or stream cabin/dash video
- Detect driver distraction signals — gaze off-road patterns, head turn persistence, attention scoring
- Draw annotations on video — bounding regions, timeline markers, event clips for QA
- Optional face anonymization on export — same pipeline can blur faces before clips leave your environment
| Layer | Where it runs | Output |
|---|---|---|
| Production ADDW | In-vehicle ECU | Real-time warning, no upload |
| Validation / fleet / insurance | Your cloud or VPC | Distraction analytics on anonymized video |
| Debug / crash review | Engineering sandbox | Faces blurred before legal or ML retrain |
Detect distracted driving. Don't ship identifiable faces unless you have a lawful basis and explicit consent.
3. Anonymize before share — BGBlur core workflow
When footage must leave the car — insurance dispute, ADAS regression, journalism, training data — bgblur.com handles:
- Face blur and face anonymization with motion tracking
- License plate blur for dashcam context
- Browser-based processing — files stay local in the client where possible
- Enterprise bulk + API for high-volume CCTV, fleet, and OEM pipelines
Workflow:
Cabin video → distraction event detected → clip extracted → faces anonymized → upload to insurer/OEM/support
Engineers still get useful signal (was the driver looking down? for how long?) without identity.
See also: how to blur video backgrounds and faces in 2026 · blur anything in video · face blur privacy guide · NY AI video disclosure law.
Who this is for
| Audience | Action |
|---|---|
| Drivers | Know ADDW ≠ mandatory police upload today — but read your OEM privacy policy |
| OEM / Tier-1 engineers | Ship closed-loop ADDW; anonymize any off-device debug path by default |
| Fleet & insurance | Score distraction on metadata + anonymized video |
| Regulators & auditors | Ask where video goes after the warning fires |
| Developers | Video Intelligence API for detection; BGBlur API for anonymization |
The bottom line
The EU didn't mandate a surveillance archive. It mandated a distraction warning — implemented in practice with cameras pointed at drivers' faces.
Safety and privacy only conflict if you design them that way. Detect attention on-device. Log events, not identities. Anonymize before upload.
That's the engineering bar — and it's what BGBlur and Video Intelligence are built for.
Next steps
- Try face anonymization free in browser →
- Video Intelligence — distraction detection & video analysis →
- Enterprise & API pipelines →
Related on explainx.ai
- EU car safety regulation 2026 — five mandatory GSR2 features (AEB, ADDW, Tesla FSD)
- Europe AI landscape 2026 — EU AI Act and sovereign compute
- EU AI Act vs US AI policy — complete guide
- Blur license plates in video — dashcam privacy guide
- Blur faces in video — AI privacy protection
- Self-host photos — Immich 3 and when to blur faces
- Flock cameras, AI surveillance, and blurring before you share
- How to blur video backgrounds and faces (2026)
Sources: Cybernews — EU driver cameras · EUR-Lex GSR 2019/2144 · Delegated Regulation 2023/2590
Regulation references accurate as of July 2026. Not legal advice — consult counsel for product-specific GDPR, type-approval, and US FMVSS/NHTSA compliance.