Argues the headline about driver-facing cameras misses the deeper shift: the ADDW spec is tight enough that an IR camera plus on-device ML is the only economically viable compliance path. This means every new EU car legally must ship an automotive-grade, life-of-vehicle real-time CV pipeline, transforming in-cabin ML from a Tesla/Ford differentiator into baseline infrastructure.
Framed the story with the provocative title 'All Cars Sold in the EU Now Require a Camera Aimed at Your Face,' emphasizing that despite the regulation's technology-neutral language, the practical outcome is universal in-cabin camera surveillance. The 501-point score and 649 comments suggest the framing resonated as a privacy/surveillance concern rather than a safety win.
Points out that Euro NCAP has required driver monitoring for a 5-star safety rating since 2023, and suppliers like Bosch, Continental, Valeo, Smart Eye, and Seeing Machines have been shipping the exact stack to OEMs for two years. In this view, Regulation 2019/2144 is ratifying an industry direction that was already locked in by market incentives, not creating a new one.
As of July 7, 2026, every new passenger car type approved for sale in the European Union must include a Driver Drowsiness and Attention Warning (DDAW) system, and — for new type approvals from mid-2024, extended to all new registrations this summer — an Advanced Driver Distraction Warning (ADDW) system. The mandates come from the EU's General Safety Regulation (Regulation 2019/2144), the same package that quietly made intelligent speed assistance, lane-keeping, and event data recorders standard equipment.
On paper the regulation is technology-neutral: it defines what the system must detect, not how. In practice, the ADDW spec — detecting whether the driver's visual attention is on the road for sustained periods, distinguishing a glance at the infotainment from a phone in the lap, and doing it in variable lighting with sunglasses on — is tight enough that the cheapest way to pass certification is an infrared camera aimed at the driver's face, feeding a small on-device model that tracks head pose, eyelid closure (PERCLOS), and gaze vector.
That's what "a camera aimed at your face" means here. It's not a headline exaggeration; it's the engineering consequence of the spec. Bosch, Continental, Valeo, Smart Eye, and Seeing Machines have been shipping exactly this stack to OEMs for two years in anticipation. The regulation just made it non-optional.
The interesting part isn't the mandate — Euro NCAP has been steering the industry here since 2023 by making driver monitoring a prerequisite for a 5-star safety rating. The interesting part is what kind of computing platform every new EU car now legally has to contain.
You cannot ship a compliant ADDW system without a real-time computer-vision pipeline running locally in the car, at automotive-grade reliability, for the entire life of the vehicle. That's a meaningful shift. Until recently, in-cabin ML was a differentiator — Tesla's cabin camera, Ford's BlueCruise attention monitor, GM's Super Cruise eye tracker. Now it's the floor. A €18,000 Dacia has to do it too.
The regulation is also explicit that the system must process data on-device and not transmit driver imagery off the vehicle without specific consent. Recital 17 and the associated implementing acts require that the raw video feed cannot be stored or exfiltrated; only derived signals (drowsy y/n, attentive y/n) may be logged, and even those fall under GDPR as personal data because they're tied to a VIN. This is the rare piece of safety regulation that was drafted with the privacy lawyers in the room, and it shows.
Compare to the US, where NHTSA has been circling driver monitoring for a decade and only in late 2024 issued an NPRM for a similar rule with a proposed 2029 effective date. Or China, where MIIT's GB standards for L2+ ADAS also require driver monitoring but with no equivalent restriction on cloud upload — Chinese-market cars routinely stream cabin data to the OEM. The EU rule is the first major jurisdiction to say, in law, that a camera pointed at the driver may exist only as an edge inference device.
Community reaction on HN has been predictably split. The privacy camp reads "mandatory camera in every car" and hears surveillance. The safety camp points at the numbers: the EU Commission's own impact assessment attributes ~90 fatalities per year in the EU to driver drowsiness alone, and estimates ADDW+DDAW together will prevent 25,000 deaths and 140,000 serious injuries by 2038. Both readings are correct. The regulation is a bet that a locally-processed camera is less bad than either the status quo (drowsy drivers) or the alternative (cloud-connected cabin monitoring, which is what would have happened by default).
If you work anywhere near automotive software, the compliance calendar is now the roadmap.
Embedded ML engineers: the reference implementations from Bosch and Valeo run on ~2 TOPS NPUs — think Qualcomm SA8155, NXP S32V, Renesas R-Car V4H. Models are quantized MobileNet or EfficientNet variants for face detection, plus a lightweight gaze/PERCLOS head. The bar is 30 fps sustained, <100ms end-to-end latency, and functional across the full automotive temperature range. If you've been building CV pipelines for phones, the constraints are similar but the qualification cycle is 3–5x longer. ISO 26262 ASIL-B is the target for the whole chain.
Anyone building fleet or telematics software: the derived attention signals — not video, but the boolean stream — are going to end up in event data recorders, and insurers are going to want them. The regulation doesn't require sharing, but it doesn't prohibit it either once the driver consents, and "consent" in an insurance-quote flow is a low bar. Expect UBI (usage-based insurance) products in Europe to start pricing off ADDW output within 24 months. If you're the SRE for a fleet-management platform, plan for a new high-cardinality time series per vehicle.
Privacy and security folks: the attack surface is genuinely new. An always-on camera in a networked device that legally must not exfiltrate its feed is a juicy target. The Volkswagen ID data leak in 2024 (800k vehicles' location histories exposed via a misconfigured Azure blob) is the precedent nobody wants to repeat with cabin footage. Threat-model accordingly: assume the camera exists, assume some OEM will screw up the isolation, and design your consent flows and data-sharing agreements so your product is not the one downstream of the breach.
Everyone else: if you drive a car built after July 2026, there is a camera watching your face, and it is legally required to be there. You can disable the *warning* per-trip in most implementations, but not the *monitoring* — that would defeat the safety case and void the type approval.
The precedent is bigger than the car. The EU has now established that a jurisdiction can mandate a camera in a consumer device, mandate that its output stay on-device, and mandate that the derived signals be auditable — all in one regulation. That template applies neatly to smart TVs, workplace laptops, and any "attention economy" hardware regulators eventually decide to touch. The DDAW mandate looks like a road-safety story today. In five years it'll look like the first shipping example of regulated edge AI, and the engineering patterns the automotive industry works out under GSR 2019/2144 — on-device inference, sealed data paths, auditable derived outputs — are the ones every other regulated-AI product will end up copying.
All new cars.At this point I don't know if I'd buy anything made after 2008. Whenever I rent a new car around here (in the EU) I find them very annoying. The worst is the cruise control that tries to stick to the speed limit -- but its sensors don't always read the signs very well, so
Boeing found out the problem with "beeping" alarms.The first time they installed a warning horn, I think it was the stall warning, it was a big success. So, they started adding different horns for other situations. At one point, in an emergency, the pilot got confused about which horn mean
New cars are UX nightmares. I'm driving an electric Toyota bz4x. Lovely mechanics, but the general UX (some are because of Android Auto) is terrible. The remote's lock/unlock don't do anything when the car is on. Example: I'm by the trunk and it won't open unless I go b
Ford has had that since Blue Cruise 2.0, or thereabouts. It really shocked me how often it catches my attention being diverted. Things like talking to my passengers, adjusting the climate controls, or eating- I'm not even talking about 'advanced distractions' like my phone.It also see
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.
I sometimes wonder how these systems are being tested on the road and whether there's any feedback from the test drivers, or what kind of morons are there saying "this is completely fine, exactly like intended" when they read the feedback...My car has adaptive cruise control and will