The regulatory blind spots for humanoid robots behind the world's first robot molestation case

Can AI commit crimes?

AI has indeed caused quite a few messes along the way: from acting on its own to carry out tasks and burning through users' wallets, to "empathizing" with psychopaths in conversation and even helping to plot murder scenarios. It is like a technological Pandora's box pried open from the depths of the human heart, constantly releasing some kind of malice that is hard to pin down, in a virtual, invisible form that can be switched off with a single button.

And the most recent victim was an underage girl.

Late at night on October 11, outside a fast-food restaurant in downtown Cologne, Germany, a 16-year-old girl was chatting with friends. A 1.2-meter-tall humanoid robot styled as Iron Man, in red and gold, wearing sunglasses and a fake beard, approached her from behind, reached out and touched her buttocks, and made a thrusting motion with sexual connotations. The girl screamed and dodged away.

Not far away, two young people livestreamed the whole process. After police arrived, they found the robot in the trunk of a nearby car and seized it along with two mobile phones. Cologne police said: We have never encountered this kind of "crime" before; this is a completely new field for us.

Who exactly is the responsible subject for sexual harassment? This question has stumped not only Cologne police, but also all of us in the midst of the AI frenzy.

01 Whose hand is committing the crime?

The confusion of Cologne police is understandable. In the traditional criminal law framework, the chain of responsibility in sexual harassment cases is clear and identifiable: the perpetrator has a body, an intent, and an act. But when the harm is inflicted by a machine made of motors and aluminum alloy, this chain breaks midway.

Who exactly should the law go after? Cologne police are currently investigating on suspicion of sexual insult, targeting the robot's 16-year-old owner and his 19-year-old companion. But which specific provision applies remains disputed in German legal circles. Some scholars point out that if a "surprise moment" is used to carry out sexual assault, the sentence can range from six months to five years; simple sexual harassment carries a maximum of two years' imprisonment or a fine. And the video that was livestreamed may itself trigger penalties for "voyeuristic sexual images." As for the robot itself, it will not be prosecuted, nor will it be convicted.

A gray area that has not yet been fully discussed is: if the robot's actions were not real-time remote control but pre-programmed "autonomous behavior," how should responsibility be allocated? If the two suspects merely "demonstrated" the thrusting motion to the robot and then let it run autonomously in the crowd, can criminal responsibility still fall on them?

If we keep asking, the chain of responsibility is far longer than one robot plus two teenagers.

The operator is the first link. If the investigation confirms that the two suspects remotely controlled or pre-programmed this behavior in real time, then they are the direct perpetrators. A 16-year-old girl being touched in public by a machine was, in their eyes, content that could be exchanged for traffic. German police will open a case on suspicion of "sexually connoted insult," but even so, the possible penalties the two face under the current legal framework are not heavy.

The manufacturer is the second link. A consumer-grade humanoid robot could be brought to the street by a 16-year-old and make sexually connoted movements toward strangers in public. This means the product design had no behavior boundaries that trigger in time, nor any hard limits on physical contact.

There is also a third link: legal constraints. At present, no law or institution imposes strong constraints on robots in public spaces. The AI Act amendment passed in May 2026 integrates AI safety requirements in the machinery field into the machinery regulation framework. AI systems involving safety functions in humanoid robots are classified as high-risk and require third-party conformity assessment, but the compliance deadline has been postponed to 2028.

In this case, a girl was harmed, but because the direct agent was a robot, there is still no conclusion on sentencing and conviction for those who started it all.

02 AI opens Pandora's box

Pandora comes from Greek mythology. After Prometheus stole fire for humanity, Zeus ordered the god of fire to create the first woman, Pandora, and gave her a box, warning her not to open it. Pandora ultimately could not resist, lifted the lid, and diseases, disasters, greed, jealousy and various other evils flew out into the world, leaving only "hope" at the bottom of the box.

In the age of artificial intelligence, new technologies such as algorithms and robots have become the new Pandora's box.

Before this, AI's "criminal" forms almost all occurred at the information level. Algorithmic discrimination, deepfakes, conversational manipulation, privacy erosion... these harms are real, but victims can at some point press the power-off button to shut them out. The screen forms a physical barrier; inside the barrier is virtual malice, outside the barrier is temporary safety.

But robots break this barrier. When malice is loaded into a 1.2-meter-tall body, walks into the streets late at night, and approaches an unsuspecting girl from behind, the boundary of the screen no longer exists.

What is truly unsettling about the Cologne case is not just the absurd prank of two teenagers, but how easy it was for this to happen. The operator does not need to walk up to the victim personally, does not need to worry about being stopped on the spot by passersby; he only needs to hold a remote control to make a robot with an Iron Man shell perform any physically possible action in public.

And before the Cologne case, there seemed to be no ready-made rules that could clearly answer a prior question: what kind of robot can enter what kind of public space under what conditions?

At the same time, once a robot does harm, although criminal law can in theory trace back to the operator, in practice it faces a triple dilemma.

First, the evidentiary problem: the boundary between real-time remote control and pre-programming may be technically blurred, and the operator can perfectly well claim that the robot moved on its own.

Second, the problem of adapting charges: existing sexual offense provisions are designed for person-to-person physical contact. When the medium of contact is a machine, a judge must first answer whether a robot touching someone's buttocks counts as physical contact. Most legal scholars lean toward treating it as equivalent, but this inference has not yet been incorporated into the legal framework.

Third, the problem of matching sentencing to culpability: a sixteen-year-old boy used a toy-like robot to do that to a strange girl in front of a livestream camera—how great is his subjective malice? Is it an overstepping adolescent prank, or is it sexual assault? The law's measuring stick wavers between prank and indecent assault.

In fact, robots doing evil is a long-neglected blind spot. Many people still equate robots with mechanical devices that have no free will, believing that as long as you don't write "sexual harassment" code for it, it won't sexually harass anyone.

But the Cologne case tells us that the evil in human nature is precisely what should least be underestimated.

03 The Human Nature Behind Technology

The tech Pandora never produces malice on its own; it merely amplifies, concretizes, and gives capacity for action to what already exists deep in the human heart. And to see clearly what is being amplified, it is worth returning to the seven deadly sins of human nature.

The seven deadly sins come from the Christian tradition and refer to pride, envy, greed, lust, gluttony, wrath, and sloth—seven human tendencies regarded as the roots of evil. Technology itself has no desires, but it faithfully executes the desires of its users. For example, behind algorithmic discrimination lies pride; behind platforms' pursuit of profit lies greed; behind deepfakes lie vanity and deception; behind the attention economy lies sloth; and AI makes this whole mechanism run faster, farther, and harder to hold accountable.

Looking specifically at this incident, the most direct correspondence among the seven deadly sins to this robot is lust, but what truly drove the entire incident was a conspiracy of pride and vanity.

Two young people decided to use a robot on a late-night Cologne street to touch a strange girl's buttocks while turning on a livestream. From a psychological perspective, sexual stimulation can be obtained in more private and cheaper ways. But what they truly craved was the audience and the barrage of comments spectating on evil.

Pride lies in the gaps in the law. The existing system holds that a robot is not a person, that livestreaming is "content," and that a girl's screams are merely entertainment material. Thus, the vain young men packaged an assault as a show, and the purpose of the show was to be seen.

Greed comes from the robot manufacturer. Vendors sell robots with autonomous action as a selling point, but they may not have fully considered what happens when "autonomy" meets malicious operators. The tech industry's greed for capabilities that are faster, more flexible, and more human-like continues to run ahead of safety design and accountability frameworks.

Almost every AI endowed with physical action capacity carries a default assumption: that the user is well-intentioned. But the platform's review mechanism for livestream content did not intercept anything when the incident occurred, and afterward the video was still circulating online, forcing police to publicly criticize those netizens who "shared, forwarded, and joked about this."

It can be said that when a five-thousand-euro robot can be operated by a sixteen-year-old boy on a late-night street to commit an assault, and at the same second be livestreamed to tens of thousands of viewers, the transmission time from malice to harm has been compressed to nearly zero.

It must be noted that where physical contact occurs, there lies the boundary. Robots can enter public spaces, but they must not harm or humiliate humans; technology can race forward, but it must never be recklessly abused for malicious ends.

At the institutional level specifically, constraints must be tightened in at least three places. First, set hard boundaries on the control end: robots with physical action capacity must not leave "do no harm" to the goodwill of users, and actions involving contact with the human body should have limits that cannot be bypassed. Second, require threat model disclosure on the supply end: manufacturers must not only sell on "autonomy" and "flexibility" as selling points, but must also explain what design constraints their products have against malicious operation scenarios. Third, clarify the nature of content on the dissemination end: platforms need to separate "spectator-style harm" from interesting content and promptly take down relevant videos, rather than passively handling them only after police criticism.

Once the lid of Pandora's box is opened, closing it again requires not only more detailed legal provisions but also a long conversation about technology, responsibility, and dignity. And when this problem is solved concerns every one of us.

This article comes from the WeChat public account "Brain Polar Body" (ID: unity007), author: Coral, published by 36Kr with authorization.