The AI safety debate has been dominated by concerns about powerful large language models, but one European VC argues that the industry is overlooking the risks of physical AI—AI embedded in robots, drones, and other hardware that interacts with the physical world. According to a Sifted article, Tapestry VC partner Audrey Miller says physical AI can cause direct harm in ways that purely digital models cannot. She argues that the current focus on LLM safety misses the need to address how AI-controlled machines might malfunction or be misused in real-world settings.
The article notes that physical AI has become a growing investment area, with Tapestry backing startups in the space. Miller calls for more attention to safety and regulation for physical AI, suggesting that the industry should treat it with the same seriousness as digital AI safety. The piece also references the broader industry push around physical AI, including Nvidia's promotion of the term, and argues that the safety debate needs to catch up with the technology's real-world deployment.
While the article does not dismiss concerns about LLMs, it makes the case that physical AI introduces a different category of risk—one that is currently under-discussed. For investors and regulators, the message is that physical AI's potential to cause physical harm demands a more proactive approach to safety, rather than waiting for incidents to force action.