Humanoid Robots and Smart Glasses Gone Wrong: Real Incidents and Ethical Problems Reported in 2026

Humanoid robots that walk, dance and work alongside people once sounded like science fiction. Smart glasses capable of recording what we see, answering questions about our surroundings and recognizing objects felt equally futuristic.

In 2026, neither technology feels particularly futuristic anymore.

Humanoid robots are appearing at public events, factories, research labs and demonstrations. AI-powered smart glasses are becoming increasingly normal consumer devices. But as these technologies leave controlled laboratories and move into everyday life, something else is becoming visible: their real-world risks are no longer theoretical.

There have already been humanoid robot malfunctions, people struck during demonstrations, privacy complaints involving smart glasses, allegations of covert recording and experiments showing how wearable cameras can be combined with facial recognition to identify complete strangers.

The interesting question is therefore no longer simply, “What could go wrong?”

Some things already have.

Humanoid robots are entering the real world — and the safety problems are coming with them

Watch a polished robotics demonstration and humanoid robots can look remarkably controlled.

They walk, balance, carry objects, perform martial-arts movements and even dance.

But robots operating in carefully prepared videos are very different from robots operating next to unpredictable humans.

A striking example appeared during a public festival in China in 2025. A humanoid robot associated with Unitree Robotics moved toward spectators behind a barrier and appeared to lunge toward the crowd, coming close to an elderly woman.

Security personnel intervened and no injuries were reported. The episode nevertheless became an early warning about what can happen when powerful mobile robots operate just a few feet away from the public.

The event also highlights an important distinction.

Calling these incidents robot “attacks” can be misleading. There is often no evidence that a robot deliberately decided to harm someone. A malfunction, loss of balance, control error, operator mistake or poorly designed safety perimeter can produce movements that look aggressive even when there is no malicious intent.

From the person standing nearby, however, the difference may not matter much.

A 35-kilogram machine swinging an arm or leg unexpectedly can still injure someone.

A humanoid robot reportedly kicked a child during a public demonstration

The risks became considerably more concrete in 2026.

On June 1, a humanoid robot believed to be a Unitree G1 was performing martial-arts-style movements during a public event in Xinjiang, China. During the demonstration, the robot struck a young boy in the abdomen.

Reports indicated that the child was not seriously injured, but the incident renewed questions about why spectators — particularly children — were able to stand within the robot’s movement range. An entry in the AI Incident Database records the case while noting that the precise cause of the movement was not established.

That’s an important point.

The ethical problem isn’t necessarily that the robot “decided” to kick someone.

It may instead be that humans placed a powerful robotic system in an environment where a predictable mechanical error could reach a child.

This changes the discussion about robot ethics.

Ethics isn’t only about whether artificial intelligence develops dangerous intentions. It’s also about deployment decisions, engineering safeguards, responsibility and acceptable risk.

Who decides that a robot is safe enough to perform two meters away from spectators?

How large should safety barriers be?

Who carries responsibility if the robot injures someone — the manufacturer, software developer, operator or event organizer?

These questions become much harder as robots become increasingly autonomous.

Losing balance can turn into a dangerous movement

Walking robots have another unusual safety problem: falling.

A traditional industrial robotic arm is normally bolted into place. A humanoid robot must continuously balance itself.

When something goes wrong, the robot may automatically perform rapid movements in an attempt to remain upright or recover from a fall.

Incident databases have documented cases in which Unitree humanoid robots reportedly made unexpected movements during demonstrations and testing. One February 2026 case involved a G1 that lost balance and allegedly struck an operator while trying to recover.

This illustrates why humanoid safety can be complicated.

A robot’s safety system cannot simply stop every motor instantly whenever something unexpected happens. Doing so could cause the entire machine to collapse.

But attempting to recover its balance can also create hazardous movements.

Engineers therefore face a difficult optimization problem: how should a robot protect itself from falling without creating a larger risk to nearby humans?

Robot safety standards are still catching up

Humanoid robots combine several technologies that previously belonged to different categories of machinery.

They are mobile robots.

They are robotic manipulators.

They contain high-torque motors.

They may use cameras and microphones.

They may navigate autonomously.

And increasingly, they can be controlled by sophisticated AI models rather than simple pre-programmed instructions.

That combination makes regulation difficult.

As of September 2026, ISO is still developing the second edition of ISO 13482, the international safety standard covering service robots. The final draft includes requirements involving human-robot contact and functional safety, but the updated standard remains in development.

Industry guidance also notes that there is not yet one single safety standard covering every possible general-purpose humanoid robot deployment. Which rules apply depends on where the robot works, what it does and how closely humans interact with it.

And humanoid robots are evolving considerably faster than international standards normally do.

That gap could become increasingly important as these machines leave demonstrations and begin working in warehouses, retail environments and eventually homes.

Smart glasses have a completely different problem: they can make surveillance invisible

Humanoid robots are easy to notice.

Smart glasses are almost the opposite.

Their appeal comes partly from looking like ordinary glasses.

Products such as Ray-Ban Meta glasses place cameras, microphones and AI inside frames that can be worn throughout the day.

For the owner, this can be extremely convenient.

For everyone standing around that owner, it creates a much more complicated question:

How do you know when you’re being recorded?

That question is now attracting attention from regulators.

In September 2026, Reuters reported that Paris prosecutors had opened a criminal investigation related to suspected sexual harassment involving smart glasses. France’s data-protection authority has also received workplace complaints and questions from organizations considering restrictions on the devices.

The concern extends beyond one case.

Smart glasses make recording less obvious than holding a smartphone toward someone. A small indicator light may signal that recording is taking place, but someone several meters away may never notice it.

That creates a fundamental consent problem.

The Harvard experiment showed how glasses could identify strangers

One of the most revealing smart-glasses experiments didn’t come from a major technology company.

It came from two Harvard students.

In 2024, AnhPhu Nguyen and Caine Ardayfio created a project called I-XRAY by combining Ray-Ban Meta smart glasses with facial-search technology, artificial intelligence and publicly available databases.

The system demonstrated something unsettling.

Someone wearing the glasses could look at another person, identify their face and potentially retrieve information such as their name, phone number, home address and relatives.

The students demonstrated the technology around Harvard but deliberately did not release the system publicly. Their stated goal was to show what existing technologies could already accomplish when connected together.

That distinction matters.

The experiment did not mean ordinary Meta glasses suddenly came equipped with a “identify everyone” button.

Instead, it demonstrated how quickly commercially available wearable hardware can become part of a much more powerful surveillance system when combined with external AI and data services.

And that’s arguably the more important ethical lesson.

The danger doesn’t always come from one product.

It can come from connecting several individually legal technologies together.

A camera.

Facial search.

An AI model.

Public databases.

Data brokers.

None necessarily looks extraordinary alone.

Combined, they can dramatically change what it means to be anonymous in public.

Smart glasses are now facing an even bigger privacy controversy

The debate intensified significantly in 2026.

Reuters reported in September that AI-powered wearable devices are facing growing international privacy concerns, with Meta’s glasses attracting particular scrutiny over covert recording, data processing and lawsuits.

Separate lawsuits reported by the Los Angeles Times involve allegations from more than 70 plaintiffs who say sensitive footage captured through Meta glasses was exposed to contractors involved in reviewing and labeling material used for AI systems.

The lawsuit alleges that some material included extremely private situations. These remain allegations being litigated rather than established findings, but they illustrate another major issue with AI wearables: people may understand that a device is recording without understanding who might eventually process that recording.

Privacy isn’t just about where the camera is pointing.

It also involves storage, cloud processing, human review, AI training, retention policies and access controls.

The biggest ethical issue may be the people who never bought the device

There is something unusual about smart-glasses privacy.

When you purchase a smartphone, you can decide whether you accept its privacy settings.

But the stranger sitting beside someone wearing AI glasses never agreed to anything.

They didn’t accept terms of service.

They didn’t configure privacy settings.

They didn’t choose the product.

Yet their face, voice or surroundings may still be captured.

This creates what could become one of the defining ethical problems of wearable AI: the device owner receives the benefits while nearby people may absorb part of the privacy cost.

A similar imbalance exists with humanoid robots.

A company may gain productivity from deploying robots, while employees, customers or bystanders are the people exposed to physical or surveillance risks.

Are humanoid robots and smart glasses inherently unethical?

No.

Both technologies have genuinely useful applications.

Humanoid robots could perform physically dangerous jobs, assist in disaster areas, handle repetitive industrial tasks and eventually help people with limited mobility.

Smart glasses can provide hands-free navigation, accessibility tools, instant translation, visual assistance and contextual information without requiring users to stare at a phone.

The ethical question isn’t whether the technologies should exist.

It’s whether their safeguards are advancing at the same speed as their capabilities.

Right now, that balance looks far from settled.

What needs to change?

Humanoid robots need stronger separation systems, safer fall-recovery behavior, clearer operating zones, reliable emergency stops and testing that reflects chaotic real-world environments rather than perfect laboratory conditions.

Public demonstrations involving children deserve especially conservative safety margins.

Smart glasses need equally serious attention to privacy by design.

Recording indicators should be difficult to disable or conceal. Users should receive clear explanations about when data leaves the device. Sensitive processing should happen locally whenever practical, and companies should minimize unnecessary data collection.

Regulators will also have to determine when wearable recording crosses the line from ordinary photography into persistent biometric surveillance.

Most importantly, responsibility cannot simply be shifted onto users.

Telling people to “use technology responsibly” isn’t enough when millions of devices are involved.

The next phase of AI ethics will be physical

For years, most debates about artificial intelligence happened on screens.

Could an AI chatbot produce misinformation?

Could algorithms discriminate?

Could generative AI replace jobs?

Humanoid robots and AI glasses add something new.

AI is beginning to acquire eyes, ears, bodies and physical access to the world around us.

A software bug in a chatbot might generate a wrong answer.

A software bug controlling a 35-kilogram humanoid robot could move an arm into someone’s face.

A privacy failure on a website might expose data you entered online.

A privacy failure involving smart glasses might capture data you never knowingly provided in the first place.

That difference is why the incidents already being reported deserve attention.

They aren’t evidence that humanoid robots or smart glasses should disappear.

They are evidence that society is moving from hypothetical AI ethics to real-world AI accountability.

And as robots become stronger and wearable AI becomes harder to notice, the next major technology debate may not be about what artificial intelligence can do.

It may be about what we are willing to let it do around people who never agreed to participate.