EU AI Act Delay: Home Robot Buyer Checklist

The better conclusion is simpler: the legal clock moved, but the physical risk did not. A robot with cameras, microphones, cloud accounts, remote support, software updates, wheels, arms, or legs still has to be judged on what it can prove today.

This article is not legal advice. It is a buyer checklist for the kind of products ui44 tracks: humanoids, companion robots, mobile manipulators, robot dogs, lawn robots, and smart-home robots that are starting to look less like appliances and more like moving AI systems.

What actually changed in the EU AI Act timeline?

The European Commission's current AI Act implementation page says the regulation entered into force on August 1, 2024, with staggered application dates. Some parts are already live: prohibited AI practices and AI literacy obligations started applying in February 2025, and general-purpose AI model rules started in August 2025.

The 2026 simplification agreement changes the high-risk schedule. The Commission says rules for systems used in certain high-risk areas, such as biometrics, critical infrastructure, education, employment, migration, asylum, and border control, will apply from December 2, 2027. For systems integrated into products such as lifts or toys, the rules will apply from August 2, 2028.

SGS's May 2026 Digital Omnibus summary adds a robot-relevant detail: machinery is being treated through a more sector-specific path. In SGS's reading, the Machinery Regulation is expected to carry much of the overlapping AI-related health and safety work for machinery. Delegated acts and guidance would then align AI requirements into existing product compliance processes.

That matters for robots because many capable home robots are not just apps with AI. They are machinery: powered systems with moving parts, sensors, control software, batteries, and failure modes that can affect human safety.

Does this mean home robots are unregulated until 2028?

No. The delay is about when specific high-risk AI obligations start applying in full. It does not erase existing product-safety, radio, privacy, consumer, cybersecurity, warranty, or market-surveillance rules. It also does not make a robot safe just because the AI Act deadline is later.

The Machinery Regulation is especially important. The European Commission says Regulation (EU) 2023/1230 applies on a mandatory basis from January 20, 2027. The Commission's machinery page says the new regulation integrates provisions for machinery with AI-powered safety functions, integrates cyber-safety for compliance-relevant software data and safety control systems, and addresses conformity assessment for machinery presenting a higher risk factor.

For a buyer, the practical question is not "which article number applies to my robot?" It is: can the manufacturer show a safety case that matches the robot's body, software, sensors, and intended use?

That is why a humanoid, a mobile manipulator, and a rolling smart display should not be judged the same way. The legal categories may still be settling, but the risk profiles are already different.

The buyer rule: ask for evidence, not labels

A manufacturer saying "AI safe," "privacy-first," or "EU ready" is not enough. A serious home robot should be able to answer narrower questions.

Start with the body. How heavy is it? How fast can it move? What force can its arms apply? What happens if it falls, loses localization, collides with a pet, or grips the wrong object?

Then look for physical safeguards. Does it have a physical emergency stop? Can it detect a person on the floor? What does it do after a failed task?

Next, ask about the AI boundary. Which functions are autonomous? Which are scripted? Which use cloud models? Which depend on remote human help? Can the robot complete a task locally if the cloud is down?

Data use belongs in the same conversation. Are camera feeds, maps, voice recordings, and manipulation logs used for training? Can the buyer opt out?

Then ask about the software lifecycle. How long will security updates last? Are firmware updates signed? Is there a rollback path? Does the company publish change logs when a behavior model changes?

Accountability matters too. If a robot becomes less safe after an update, who is responsible?

Finally, ask about support. Where is the EU repair route? How are batteries, actuators, wheels, arms, sensors, and covers replaced? Is there a local service partner, a mail-in path, or only remote chat support? A robot that cannot be repaired safely is not ready for homes, no matter how polished the demo looks.

What ui44's database says about the gap

ui44 currently tracks 283 robots. The categories most likely to raise physical-AI compliance questions are already large: 81 humanoids, 36 companion robots, 14 home assistants, and 11 quadrupeds. Those are not all consumer products, but they show where the market is moving.

Connectivity is nearly universal. In the database, 254 robots have explicit connectivity entries, including 167 with Wi-Fi, 85 with Bluetooth, and 43 with Ethernet or LAN. That does not mean every connected robot is risky. It means cybersecurity is no longer optional. A moving robot is a networked computer with a body.

The certification picture is less complete. Only 65 of 283 robots currently list at least one certification or formal rating in ui44's structured data. That number includes everything from CE and FCC marks to IP ratings, medical-device status, privacy certifications, and product-specific awards. The point is not that every robot lacks compliance. The point is that public buyer-facing disclosure is uneven.

That disclosure gap matters more as robots move from novelty to household infrastructure. If a company wants you to trust a robot near children, pets, older adults, medications, stairs, blades, doors, or kitchens, it should publish more than a teaser video.

How this applies to real home robots

The EU AI Act delay looks abstract until you map it onto actual products.

1X NEO is one of the clearest home-branded humanoids in the database. ui44 lists it as a $20,000 preorder, 167 cm tall, 30 kg, with about 4 hours of battery life, Wi-Fi, Bluetooth, RGB cameras, depth sensors, tactile skin, microphones, and household-chore ambitions. That is exactly the kind of robot where buyers should ask for a safety case, remote-help policy, data policy, and support plan before treating the preorder as a normal appliance.

Unitree G1 is different. It is available, starts at $13,500, and is positioned more like an affordable humanoid research and developer platform than a plug-and-play home helper. ui44 lists Wi-Fi 6, Bluetooth 5.2, a depth camera, 3D LiDAR, a microphone array, ROS 2 compatibility, and about 2 hours of battery life. That makes it concrete hardware, but not a shortcut around buyer due diligence. A research humanoid in a home still needs clear limits.

Hello Robot Stretch 4 is a useful contrast. It is not a bipedal humanoid. It is a wheeled mobile manipulator priced at $29,950, available now, with a 160 cm working height, a 45 cm footprint, 8 hours of light-load runtime, and a 2.5 kg extended-arm payload. Its shape narrows the risk: no legs, one arm, a clearer assistive-pilot story, and open ROS 2 / Python software. That does not eliminate compliance questions, but it makes them more specific.

Samsung Ballie sits at the other end of the evidence spectrum. It has a compelling smart-home story: a rolling companion, projector, SmartThings control, cameras, sensors, Gemini, and Bixby. But ui44 still lists no public price, no firm release date, and many undisclosed hardware basics. Under a buyer checklist, Ballie is not "bad"; it is not yet evidenced enough.

Cybersecurity is becoming part of robot safety

The most useful May 2026 signal may not be the deadline delay. It may be an industrial certification story. SGS says DOBOT's CR 30H collaborative-robot series achieved ISO 10218-1:2025 cybersecurity compliance after assessment of threat modeling, access control and identity authentication, secure communication, configuration protection, port and interface management, and secure software updates.

That is not a home robot certification, and it should not be treated as one. But it shows the direction buyers should expect. Cybersecurity is no longer separate from safety when a machine can move, sense, and receive commands. An insecure update channel or remote diagnostic path can become a physical safety problem.

For home robots, the same checklist should be normal:

This applies beyond humanoids. Enabot EBO X is a 1.5 kg companion robot with 2.4 GHz and 5 GHz Wi-Fi, a 4K stabilized camera, far-field microphones, Alexa compatibility, and GPT-4o mini integration in ui44's data. Yarbo M is a modular yard robot with RTK, optional LiDAR/camera sensing, Wi-Fi, Bluetooth, HaLow, optional 4G, blades, snow and leaf modules, and a preorder path. Those are very different products, but both show why connected-robot security is a buyer issue, not an enterprise-only issue.

Should EU buyers wait until 2028?

Not automatically. Waiting makes sense if the product is vague, expensive, physically powerful, heavily cloud-dependent, or missing support details. It also makes sense if you are importing a robot whose EU compliance path is unclear.

But the better rule is not "wait for the law." It is buy only when the evidence matches the risk.

A small companion robot with no arms and a clear privacy policy is a different risk than a 35 kg humanoid with dynamic walking. A wheeled manipulator deployed in a supervised assistive pilot is different from a consumer robot sold as an autonomous household worker. A robot mower with blades, RTK, cameras, and remote diagnostics needs a different checklist again.

Use the delay as leverage. If a manufacturer says the rules are not fully in force yet, ask what standards, audits, tests, service plans, and incident processes it is already using. Serious companies should welcome that question.

Bottom line

The EU AI Act delay gives robot makers more breathing room. It does not give buyers a reason to lower their standards.

Between now and August 2028, the smart move is to separate robot demos from robot evidence. Ask how the machine handles physical safety, cybersecurity, human oversight, updates, repair, incident reporting, and data use. Compare those answers against concrete specs in the ui44 robot database, then use the compare tool to see which robots disclose enough to trust.

For home robots, the future deadline matters. The robot in your hallway matters more.