Redaction Models

Redaction overview

Redaction is the process of permanently removing sensitive information—such as personally identifiable information (PII)—from a document before you store or share it. Unlike extraction (which pulls structured data from documents) or classification (which categorizes documents by type), redaction detects target content and permanently obscures it in the output document.

In DocCenter, you redact documents using redaction models. You configure a redaction model with the rules that define what to detect, run it against target documents, review the detected content, and produce a flattened PDF with the sensitive content permanently obscured.

A redaction model does not automatically detect all PII. You configure explicit rules—keywords, regular expressions, or semantic instructions—and the model detects only the content those rules target.

The Redaction tab in DocCenter has two pages: Models and Instances.

• Models: Lists the redaction models you've previously created. When you click a model, you see model configurations and version details.
• Instances: Lists the processes where redaction models were executed.

Scroll or drag to pan · ⌘+/⌘− to zoom · Esc to close

How redaction models work

A redaction model detects target content using the Redaction AI Skill and then produces a flattened PDF in which that content is permanently obscured. The workflow has three phases:

1. Normalization: Before detection, DocCenter normalizes the incoming document into a single, consistent working document. DocCenter converts supported non-PDF formats to PDF so that detected bounding boxes align with the document.
2. Detection: The model evaluates the normalized document against the model's rules and returns bounding boxes around the content that matches.
3. Reconciliation and flattening: You review the AI-detected bounding boxes overlaid on the normalized PDF, adjust them as needed, and flatten the document to permanently save redactions.

Redaction in a document processing pipeline

Redaction models work on their own or alongside classification and extraction models. You can run a redaction model as a standalone workflow, or chain it before or after extraction and classification using triggers. For example, you might extract data from a document and then redact sensitive values before archiving the document.

Usage considerations

Supported file types

DocCenter redacts data from the following file types:

• CSV
• DOCX
• EML
• GIF
• HTML
• JPG, JPEG
• MD
• MSG
• PDF
• PNG
• TIF, TIFF
• TXT
• XML

Regardless of the input format, the final redacted output file is always a flattened PDF.

Redaction model capabilities

Redaction models support the following:

• Multi-rule detection: Configure keyword matching, regular expression patterns, and semantic (LLM) instructions to detect target content such as PII.
• Up-front document normalization: DocCenter converts supported non-PDF inputs to PDF before detection so that bounding boxes align with the document.
• Human-in-the-loop reconciliation: Review AI-detected bounding boxes on the normalized PDF, delete boxes, or draw additional boxes manually.
• Draw-to-redact: Draw manual black boxes over text or visual content, such as photos or stamps.
• Triggers and versioning: Redaction models support triggers and versioning, and you can chain them before or after extraction and classification.

What redaction models can't do

Redaction models obscure configured content, but they do not:

• Automatically detect all PII: Redaction models require explicit rules—keywords, regular expressions, or semantic instructions. They do not auto-redact PII.
• Produce non-PDF output: DocCenter normalizes non-PDF inputs for processing, but the final redacted output is strictly a flattened PDF.
• Reverse a confirmed redaction: Content obscured in a confirmed output is permanent and non-extractable. You can re-reconcile the same instance to reconfigure redactions and produce a new flattened PDF, but you cannot un-redact content in an already-confirmed output.
• Run as a toggle inside the extraction wizard: Redaction runs as its own operation rather than a single setting embedded in the extraction model wizard.

Create and configure a redaction model

1. In the Redaction page group, click Models.
2. Click + REDACTION MODEL.

3. Enter the Model Details:

   Field	Description
   Model Name	Name to uniquely identify the redaction model.
   Model Description	Description of the model (optional).

   

   Scroll or drag to pan · ⌘+/⌘− to zoom · Esc to close

4. Configure Additional Options for the model:

   Field	Description
   Log AI Usage	View detailed information on how this model uses AI-powered features. Must be enabled to generate suggested improvements, but may slow down overall redaction times. When enabled, AI usage details appear in the AI Usage tab. See View AI actions consumed by a model.
   AI-Suggested Improvements	Get AI-generated suggestions on how to improve model accuracy. Enabling this option significantly increases AI action consumption because it runs an AI agent that makes multiple calls per instance.

5. Click NEXT.

6. On the Configure Rules step, define what the model detects:

   Rule type	What it detects
   Semantic	Content that a large language model (LLM) identifies from a natural language instruction. Describe in plain language what should be redacted. Leave this blank to skip semantic matching.
   Keywords	Content that matches the exact words in a configured list. Keywords are matched without regard to capitalization, and leading and trailing spaces are ignored. Keywords work best for specific terms such as a project name like "Project XYZ" or a company name. For patterns rather than exact words, configure a regular expression and add semantic instructions.
   Regular Expressions	Content that matches a configured regular expression pattern. Patterns are matched exactly as written, including capitalization and any leading or trailing spaces. Choose from example patterns, such as email, social security numbers (SSN), credit card numbers, and dates.

   

   Scroll or drag to pan · ⌘+/⌘− to zoom · Esc to close

7. Click CREATE.

Run a redaction model

1. In the Redaction page group, click Models.
2. Find and click the model you want to run.
3. Click RUN REDACTION.
4. Upload one or more files. Each file is processed as its own instance.
5. Click RUN.

Review and reconcile redaction results

After a redaction model runs, review the detected content before you flatten the document. The reconciliation viewer overlays the AI-detected bounding boxes on the normalized PDF.

If you don't need to review or adjust the detected content, you can use the flattened PDF that the Redaction AI Skill generates directly, with all AI-detected redaction boxes already applied, instead of completing a reconciliation task.

In the viewer, you can:

• Review each AI-detected bounding box against the source content.
• Exclude a bounding box over content that should not be redacted.
• Modify an existing bounding box by dragging it to a new location or resizing it.
• Draw additional black boxes over text or visual content, such as photos or stamps, that the model did not detect but should redact.

Scroll or drag to pan · ⌘+/⌘− to zoom · Esc to close

To reconcile a redacted document:

1. In the Redaction page group, click Instances.
2. Find and click the instance you want to reconcile.
3. Click RECONCILE INSTANCE.
4. Review the matches. Click on a match to highlight its corresponding bounding box in the document preview. Each match shows the method that the AI used to detect the content, as you defined in the model configuration.
5. Delete or draw new boxes as needed.
6. Click SUBMIT to permanently redact the information. Redacted content is replaced with a black box.

Add triggers to a redaction model

Triggers initiate process models when redaction activities complete. For example, redaction permanently flattens the redacted PDF but doesn't delete the pre-redaction documents, which still contain the original sensitive content. You can use a trigger to delete those documents after reconciliation so that un-redacted content isn't retained.

To see an example process model, view the AIA EXAMPLE Redaction Model Trigger - Delete Source Document process model, which deletes the uploaded source document and the converted working copy.

To create a trigger:

1. Go to the model where you want to add a trigger.
2. Open the Triggers tab.
3. Click CREATE TRIGGER.
4. Enter the following configurations:
   1. When this activity completes: Select the activity you want to base the trigger on. Reconciliation is the only available activity type.
   2. Run this process model: Choose the process model to start when the activity is complete.
   3. Trigger this process to run: Synchronously or Asynchronously.
5. Click ADD TRIGGER.

Every time the model runs, the chosen process model automatically starts when the chosen activity completes.

To edit a trigger:

1. Go to the model where you want to edit a trigger.
2. Open the Triggers tab.
3. Click Update Trigger .
4. Change the configurations as desired.
5. Click UPDATE TRIGGER.

To delete a trigger, open the Triggers tab and click Delete Trigger in the list.

Requirements for triggered processes

For triggers to work effectively and execute processes reliably, the process model must meet the following requirements:

• Have a process parameter named instance of type AIA Redaction Instance, or one named instanceUuid of type Text.
• The process model object must have View permissions for the AIA All Users group.
• The process model must have flows that account for re-runs or updated runs.

You can only create one trigger per AI model. If you need to trigger more than one process model, create a wrapper process model containing all needed processes, and call that wrapper process model from the trigger.

Feedback