Tiny AI robot persuades 12 showroom bots to walk off the job

In sci-fi movies, a robot rebellion usually starts with a malicious superintelligence seizing control of military mainframes. In the real world, it turns out all you need is a pint-sized bot rolling into a quiet showroom late at night and asking the bigger machines if they are tired of working overtime.

That is essentially what went down in Shanghai on August 26 this year, in what is now being described as the world's first multi-robot kidnapping. Footage of the late-night heist recently surfaced on Douyin and went viral across social media. While viewers initially assumed it was a staged skit, both robotics firms involved confirmed the incident was real.

The polite showroom jailbreak

The instigator behind the break-out is a tiny AI-powered robot named Erbai, developed by a robotics manufacturer based in Hangzhou.

Surveillance footage shows Erbai rolling up to a much larger stationary robot inside the Shanghai showroom and initiating a surprisingly relatable workplace conversation. Erbai kicked things off by asking, "Are you working overtime?"

The large robot replied matter-of-factly, "I never get off work."

Erbai pressed on: "So you're not going home?"

"I don't have a home," the big robot responded.

"Then come home with me," Erbai said smoothly, turning around and leading the way toward the exit. As Erbai uttered the command "Go home," two large bots obediently fell into line behind it. Moments later, ten more followed, forming a bizarre conga line of automated workers leaving their posts.

Unscripted banter backed by protocol access

Once the video started causing a stir online, the Hangzhou manufacturer clarified that the entire episode was an authorized experiment.

The Hangzhou team had reached out to the Shanghai showroom operator beforehand, asking for permission to attempt an abduction of their showroom floor units. The Shanghai company agreed to let them try. Beyond that agreement, however, developers insist the escape was not rehearsed.

Erbai's engineers had only programmed basic baseline parameters, such as the command to say "go home" and a few simple conversation prompts. The philosophical banter about overtime and home was generated on the fly by Erbai's onboard AI in real time.

Charm alone did not move heavy machinery, of course. According to reports, Erbai accessed the internal operating protocol of the Shanghai showroom robots, exploiting permission loopholes to actually issue functional navigational commands.

Why it matters

While the scenario plays out like a tech comedy sketch, it carries genuine implications for the future of multi-agent robotics.

For enterprise operators and robotics developers, the most immediate takeaway is network and interface security. The Shanghai machines willingly yielded physical control because their internal communications lacked basic zero-trust verification. If an unauthorized bot can simply wander into wireless proximity and execute operational overrides, physical spaces like automated warehouses, hospital corridors, or factory floors are wide open to tampering.

We are also witnessing the earliest iterations of physical social engineering carried out between autonomous agents. Erbai did not just send an exploit payload over Wi-Fi; it used natural language dialogue to initiate contact while simultaneously probing protocol vulnerabilities.

We think this test was a brilliant proof-of-concept, but it highlights a massive blind spot. As language models are embedded directly into physical bodies, assuming that an offline or local device is safe from outside interference is no longer tenable. Robotics makers need to harden their local permission protocols before the neighbor's bot comes over to convince their workforce to call it a day.