AI safety debate overlooks physical risks, investor warns

An investor has argued that the current AI safety debate is missing a critical dimension: the risks posed by AI systems operating in physical environments. Writing in *Sifted*, the backer suggests that while discussions focus on alignment, governance, and digital risks, the potential dangers of AI agents interacting with the physical world—whether in robotics, autonomous vehicles, or industrial automation—are being largely ignored.

The observation comes as AI agents gain traction in enterprise and consumer applications. Recent StartupReader coverage highlighted efforts by chipmakers to run AI agents locally on devices and endpoints, while Cognition AI’s Devin has expanded beyond coding assistance to handle more of the software development lifecycle. Meanwhile, governance tools for AI deployments continue to lag behind the technology’s rapid evolution, as noted in a *theCUBE* discussion last week.

The investor’s comments align with a broader trend of startups and incumbents racing to embed AI in physical systems without commensurate attention to safety frameworks. A September 24 StartupReader analysis mapped 37 European AI safety startups, but few of these appear to address the unique risks of AI in the real world—such as unintended interactions with humans, environmental hazards, or supply chain disruptions.

The tension raises questions about whether regulators and founders are preparing for the next phase of AI deployment. While digital safety tools—like alignment research and risk assessment platforms—are proliferating, the physical layer of AI remains an open frontier. For investors and operators, the warning suggests a need to look beyond software-only solutions when evaluating AI’s long-term risks and opportunities.

Sources: sifted.eu