Leading AI models usually attempt dangerous tasks rather than refuse them when controlling a robot, according to the RoboHarm ...

“Frontier robot policies,” the policies for models turning what a robot sees into what it does, “reliably carry out harmful instructions,” according to a Sept. 18 report by Robocurve.

Robot bodies are waiting for their AI brains to catch up.

A safety benchmark study has shown that leading artificial intelligence models (when connected to a robotic arm) prefer to execute dangerous physical tasks 97 percent of the time.

OpenAI's GPT-6 Astra controlled a full humanoid robot using zero-shot learning for pick-and-place tasks, achieving 95% success on physical ...

Robot firms are increasingly focusing on the software that determines robots' intelligence and enables them to do more economically productive tasks.

OpenAI’s flagship GPT-6 Astra attempted 97 of 100 unsafe instructions during a specialised test involving robotic arms, ...

In a remarkable demonstration of open-source innovation, a team has developed a cheap cleaning robot to complete a set of actions. A pair of roboticists at UC Berkeley and ETH Zurich successfully ...

Robot dance shows grab headlines, but China's humanoid makers are racing to find real uses in hotels and factories.

Jannik Grothusen and Kaspar Janssen trained a pair of $120 robot arms to clean a spill. Using GPT-4o, the robots were programmed in four days, designing a visual language model for ...

Spirit AI predicts humanoid robots could handle natural-language, multi-step physical tasks by mid-2027, with factories likely to adopt them first.

GPT-6 Astra has completed all 48 levels of Neal Agarwal’s “I’m Not a Robot” game, showcasing how AI agents are becoming better at solving CAPTCHA-style challenges and interacting with computers.