Systems in this category ingest a facility's existing telephone and messaging infrastructure, transcribe calls using automatic speech recognition, and apply keyword search and increasingly semantic or agentic AI analysis to identify coded language, relationships between callers, and possible security threats. Transcripts become searchable by keyword and other metadata, and keywords are typically selected jointly by facility officials and the vendor.

The most documented example, Verus from LEO Technologies, has been deployed across state prison systems and county jails in at least eight US states, monitoring tens of millions of calls a year. An investigation obtained contracts and emails showing the system used well beyond its stated gang-and-violence rationale in practice, including flagging calls about matters unrelated to any security concern.

What distinguishes this category from retrospective analysis of a seized device is that it operates on live telephone infrastructure continuously and at population scale, and it necessarily captures the other party to every call -- typically a family member who has not been convicted of anything and has not consented beyond the standard notice that calls are monitored. Newer generations of these systems have extended from call content into continuous voice-identity tracking (detecting when a caller's voice does not match the profile associated with the phone credential being used) and into AI video analysis of the facility itself.