The moral buffer: humans acting through robots

A humanoid robot allegedly touched a teenage girl inappropriately outside a Cologne fast-food restaurant while two people filmed, livestreamed, and provided commentary. In San Francisco, a YouTuber got knocked down in a cage fight against three humanoid robots. Both times the machines made the headlines. The people making choices around them faded into the background.

On Saturday night, 10 October, a red-and-gold robot about 120 centimetres tall allegedly touched a 16-year-old girl without consent outside a fast-food restaurant in Cologne. Police name the robot’s 16-year-old owner and his 19-year-old companion as suspects who streamed and narrated the incident on social media. The investigation concerns a suspected offence described as “insult on a sexual basis”; no charges have been announced. I checked the Cologne police release: its headline calls the robot presumably remote-controlled, but investigators are still checking how it moved and whether anyone deliberately steered it. PC-WELT is more definite about remote control than the police account and says the suspects controlled the robot remotely.

IT-law and robot-ethics expert Iris Phan says the robot is a tool, not the perpetrator. She sees little evidence of an autonomy problem and much to suggest an incident staged for an audience. She calls the robot a “moral buffer”, a way to create distance from the victim and lower inhibitions. Mohamad El-Ghazi, a criminal-law professor at the University of Trier, says “it makes no legal difference whether contact is made with a stick, a broom handle or a robot”. Both experts see no gap in current criminal law. Other offences could apply, depending on the evidence. The law already knows how to handle people using tools.

San Francisco offers the same basic arrangement in a different setting, with a cage and an audience. On 18 September, Robot Entertainment Kombat (REK) staged a cage fight between YouTuber Frankie LaPenna and three humanoid robots. LifeZette identifies them as a Unitree G1 and two modified EngineAI T800s. REK’s technology partner says human pilots controlled the machines remotely in real time, using VR gear or a gamepad. LifeZette also reports that LaPenna was knocked down and injured his hand. On 30 September, Cix Liv (the founder of REK) posted a cease-and-desist letter from California’s State Athletic Commission. It ordered REK to stop "ALL PERSONS HOLDING, PROMOTING, ADVERTISING OR SPONSORING UNSANCTIONED HUMAN VERSUS HUMANOID ROBOT CAGE FIGHT" and warned that unlicensed events are a misdemeanour. The letter leaves robot-versus-robot fights outside the dispute. The commission used existing combat-sports licensing rules, not a rule written for robots. After a separate robot-versus-robot match, Liv confirmed: “We're still a way out from autonomous fighting.”

​​​​​​In Cologne, deliberate control remains unconfirmed. At REK, remote piloting was the point. The public spent years worried about robots acting against people. These two cases are people acting through one. As Phan put it: a body you can drive from a distance makes it cheaper to do what you would not do in person.

Picture a humanoid losing its balance and falling towards someone standing in front of it. Has it been tested for that?

AGIBOT says its A3 Ultra received safety certifications from TÜV Rheinland. The certifications cover rules for household appliances and general machinery. I read the certificates from Certipedia, but couldn't find any record whether the robot was tested for falls, balance, or working near people.

Their A2, X2, A3 and G2 are all certified as a household appliance set, tested at one point in time, against rules written for machines that do not move and a general method for assessing machinery risks. Those rules serve a purpose. But they weren't designed to answer some very basic questions about a humanoid. What happens when it loses its balance? How safe is it when it falls near someone? What happens when a person steps into its path? The certificate doesn't answer those questions. The records show no fall testing, balance testing or testing of the robot working near people. Those tests may have happened, but the paperwork being presented as evidence of safety doesn't tell us whether they did.

The other certificates cover different things. One deals with radio signals and electromagnetic interference and covers whether the robot uses radio frequencies properly and doesn't interfere with nearby electronics. The records also include a voluntary cybersecurity license, with checks on the device cybersecurity standard like default passwords, update mechanisms, credential storage, although this license was not part of AGIBOT's announcement.

So here's the problem: the industry doesn't yet have a shared rulebook for testing humanoids as complete machines. TÜV SÜD says no shared European standard covers humanoids as a whole. An international standard for robots that actively balance themselves is still being developed. China kicked off formalising its own humanoid testing standards this August.

Testing bodies are left adapting rules written for other machines, while humanoids are already moving into workplaces where people share their space. It’s the same problem I raised with emergency stops: what keeps a walking robot safe isn’t always what keeps an ordinary machine safe. The robots have arrived while the rulebook is still catching up.

This week's Signal covers two incidents that shared a headline with the robots, and not much with the humans behind them. Below the Fold: the safety certification on a humanoid robot doesn't tell you whether it's been tested for falling on someone.

Both pieces this week are about robots in spaces where the rules weren't written for them. Criminal law already knows what to do with a tool used to harm someone. Safety certification doesn't. The tests weren't designed for a machine that walks, balances, and falls near people.