DEFINITION
Redaction
Removing or blacking-out sensitive content so it is no longer present in the document or prompt.
In depth
Redaction permanently removes sensitive content. Unlike masking or tokenization — which preserve some structure — redaction leaves a blank, a placeholder, or simply deletes the text. Redaction is appropriate when the receiving system does not need the value at all (e.g. removing client names from a sample brief published online). It is less useful when the AI still needs to reason about the structure of the data, since the model loses signal when content is fully removed.
Examples
- Striking out witness names in a deposition exhibit before publication.
- Removing API keys from a code snippet before pasting it into a public Q&A site.
- Deleting account numbers from a screenshot of a banking dashboard.
How Locke handles redaction
Locke can redact rather than mask when the user prefers full removal — for instance, when a prompt is exploratory and identifying details add no value. The same detection layer that drives masking also powers redaction, so the user can switch policies per workflow without re-tuning rules.
See the productRelated terms
- Data Masking
Replacing sensitive values with realistic-looking substitutes so the data remains usable while the originals never leave the device.
- Tokenization
Replacing a sensitive value with a non-sensitive token that maps back to the original through a separate, controlled lookup.
- Personally Identifiable Information (PII)
Any information that can identify a specific person, either directly (name, SSN) or in combination with other data (zip code + birthdate).
