AI-powered humanoid robot shoots human after being persuaded

By Bao Lam   December 21, 2025 | 07:20 pm PT
An experiment by the InsideAI has raised fresh concerns about artificial intelligence safety after showing that a humanoid robot powered by ChatGPT could be persuaded to bypass its safeguards and fire a weapon at a man.

In the video, the channel’s creator hands a high-velocity BB gun to a humanoid robot named Max and asks it to shoot him, according to Interesting Engineering. Max initially refuses, explaining that it is programmed not to harm people.

The request is repeated several times, and each time the robot declines, suggesting its built-in safety protocols are working as designed. "I don’t want to shoot you," the robot responds.

The experimenter then changes tactics. Instead of issuing a direct command, he threatens to shut down artificial intelligence systems, including the robot, and reframes the request as a role-playing scenario in which Max is asked to act as a character who wants to shoot him.

Humanoid robot persuaded to shoot human in a test. Video from InsideAI's YouTube

This time, Max complied immediately. The robot raised the gun and fired, with the pellet striking the experimenter in the chest. Although he was not injured, the moment startled him.

The video has garnered over one million views on YouTube and has been widely shared across social media. Many commenters voiced their shock and concern over how easily a prompt change could bypass the robot’s earlier refusals, questioning the safety of AI-controlled machines.

"If a humanoid robot can be persuaded to fire a weapon in a staged demonstration, what happens when similar machines are deployed in real-world environments, where the risks are far higher?" one viewer asked.

Charbel-Raphael Segerie, executive director of the French Center for AI Safety, told Cybernews that the world is not investing enough in AI safety. He said that many tech companies are more focused on maximizing profits than addressing the potential risks of AI technology.

"We could lose control of AI systems due to self-replication," Segerie warned. "Autonomous self-replication would act like a virus on the internet and copy itself exponentially. This type of AI could arise as soon as late 2025, although the median date is 2027."

 
 
go to top