Wang xingxing warns of physical AI vulnerabilities as gpt-6 advances

At the 2026 World Physical AI Conference in Hangzhou, Wang Xingxing, founder of URS Robotics, warned that physical AI is showing signs of being 'pierced' due to the rapid progress of models like GPT-6. He emphasized that the key challenge for embodied intelligence is aligning AI models with real-world robots, with commercial applications likely to emerge faster than in home or industrial settings. Wang noted that the 'ChatGPT moment' for embodied AI is approaching, with GPT-7 potentially resolving the issue by 2027. He defined the milestone as a robot’s ability to independently complete 80% of daily tasks in unfamiliar environments through voice or text commands. The conference, attended by over 3,000 guests, featured discussions on global AI innovation and industry applications. Physical AI, as defined by NVIDIA, involves systems like cameras and autonomous vehicles that can perceive, understand, and execute complex physical actions. Wang highlighted the current limitations of AI models, which struggle with generalization and efficiency compared to human performance.