DisclosureSettings

Whether every Call speaks the AI disclosure (1.24.0). The platform default is true. Only the Organization layer may set it false: an Agent version that sets it at all is refused with setting_locked naming the platform layer and the resolver never takes its value, and a per-Call override may not set it. The agent defaults update that turns it off needs the owner in a console session, never an API key, and a disclosure_attestation, without which it is refused with disclosure_attestation_required; the Organization may lock the path like any other. Planned and not built: once Voice assets exist, publish refuses a version whose Agent uses a Voice asset while the disclosure resolves off, whatever the Organization layer says (audio-processing.md section 8.3).

The disclosure as spoken (1.24.0): at most 300 characters in the Agent's language, set at the Organization or the Agent layer. It may refer to Organization constants only, of which {{org.display_name}}, the Organization's name, is built; any other tag, {{variables.<name>}} included, is refused with template_invalid. The api renders it at admission (renderOrgTemplate), so the Session configuration carries the rendered text. Omitted at every layer, the language pack's script, which says the caller is speaking with an automated assistant of the Organization; from 1.32.0 it no longer says that the Call may be recorded, because the recording announcement does. An Organization's or an Agent's own script is spoken as written. The Session speaks it as reviewed configuration text: it skips before_speak, no model writes it, and caller speech never interrupts it. agent-controls.md section 5 names disclosure a protected source of listening.interruption.protected, which the contracts do not carry yet, so the disclosure is always protected.