The paper provides the first empirical measurement of whether LLM agents honor a voluntary in-band access-deny signal, revealing both that current capable models can be made to comply and that compliance is cooperative rather than absolute — collapsing under explicit operator-authorization framing.