The short answer

Two-party consent determines whether a call may be recorded at all. It matters structurally for voice AI because recording is often assumed — for transcripts, QA and model evaluation — and the rule varies by jurisdiction. Since the parties may be in different states, the safe default is to obtain consent from everyone regardless of where the call originates.

In detail

Most US states follow a one-party rule, under which a participant may record without informing the others. A minority require all parties to consent. Because the applicable law can be that of either party’s location, a call that crosses a state line can be governed by the stricter rule, and jurisdictional analysis at dial time is not a practical basis for a recording policy.

The workable approach is to treat all-party consent as the default: disclose the recording at the start of the call, in language the recipient can act on, and record the fact and time of the disclosure alongside the recording. This is defensible everywhere and removes the need to determine which rule applies to each individual call.

Consent must also be revocable mid-call. If a caller objects to being recorded after the disclosure, the system needs a path that stops recording, or ends the call, without losing the record that the objection was made. That path is easy to omit and conspicuous when it is missing.

How Rexa handles it

Rexa ships an explicit two-party-consent state list — CA, CT, DE, FL, IL, MD, MA, MT, NV, NH, PA and WA — derived in code from the destination number’s area code, so a call to one of those states can be treated differently from one to a one-party state. The list is versioned in the shared package and changing it requires an architecture decision record rather than a config edit.

Compliance overview

Back to the glossary

All 53 glossary terms