We write about why AI agent sandboxes are insufficient. Identity, not just policy, is critical to securing autonomous AI systems and preventing misuse.