Taiwan’s AI-assisted cyberattack exposed a hard truth about self-hosted agents: local control does not equal safe control. The real risk is what your agent can reach after one bad instruction. Read the