A transcript is the written record of a spoken exchange, such as a voice prompt, call, or chat session. In security systems, transcripts are easier to search, index, retain, and analyze than raw audio, which makes them useful for logging, moderation, and model training. That same convenience also increases risk: a transcript can preserve sensitive commands, personal data, authentication details, or regulated content long after the conversation ends.
Transcripts matter because they turn a temporary voice interaction into durable data. Attackers may target them through account compromise, weak access controls, or overbroad retention policies, then mine the text for secrets or intelligence. Defenders should treat transcripts like any other sensitive record: limit retention, restrict access, support deletion, and separate consented training data from non-consenting user content. In voice systems, the privacy question is often not what was heard, but what the transcript makes permanently searchable.