Will Anthropic’s AI Safety Crisis Hit Humanoid Robotics?

🚨 AI safety is no longer just a software problem. And the @Anthropic safety saga could well extend into the world of humanoid robotics very soon. In this episode of @overthehorizon, I’m joined by Scott Walter and Gustav Andersson for a wide-ranging Sunday Bot Roundup on what happens when powerful AI systems move from screens and servers into humanoid robots, factories, homes and the physical world. We start with the latest Figure.03 humanoid robot stair-climbing demos and what they reveal about robot perception, balance, foot placement, actuator wear, shock absorption and the difficulty of testing robots in real-world environments. Then we dig into the Anthropic AI safety saga and what it could mean for embodied AI. If powerful AI models raise concerns in software, what happens when similar capabilities are placed inside humanoid robots that can move, manipulate objects and interact with people in homes, workplaces and public spaces? We also discuss OpenAI’s return to robotics, the rise of world models, and why humanoid robots may need a deeper understanding of physics, objects and human environments before they can become truly useful. Scott and Gustav weigh in on the debate around Unitree’s dominance in humanoid robotics. Is Unitree really building an insurmountable lead, or is the company still facing major questions around dexterity, payload, overheating, durability, use cases and real deployment? We also look at Weave Robotics’ Isaac home robot, and why folding laundry, picking up objects, moving around homes, avoiding pets, handling stairs and even vacuuming are far harder than they look. Plus, we cover a Unitree robot being prepared for a Mount Everest challenge, Xiaomi CyberOne robots entering factory internships, and autonomous farm robots using UV light for chemical-free pest control. This is a reality check on the next phase of humanoid robotics: not just better demos, but real-world safety, reliability, autonomy and usefulness. #OverTheHorizon #OverTheHorizonPodcast #HumanoidRobots #AISafety #Anthropic #Unitree #FigureAI #Robotics