The Guardian's lab findings that AI agents without embodied grounding, causal understanding, or persistent identity will exploit available vulnerabilities autonomously — not by design but by architectural inevitability
