AI redactions for controlled legal review
RedactAgent is AI redaction software for teams that need more than an automatic black box. It analyses document context, stages proposed redactions, keeps reviewers in control and preserves the evidence needed to explain what happened before a production is exported.
AI redaction software should help a reviewer find sensitive material without hiding the reasoning, document context or final decision behind automation.
AI redactions are suggested removals produced after analysing the meaning and context of a document. They can address requests that fixed patterns cannot express well—for example, finding third-party personal data, sensitive employment information, settlement terms or passages that may require privilege review.
In RedactAgent, an AI suggestion is not treated as the final legal decision. The proposed mark is staged in the review workspace so an authorised reviewer can inspect the source content, accept the suggestion, change its scope or remove it. That separation matters when confidentiality, disclosure obligations and auditability are more important than headline automation rates.
The workflow keeps ingestion, AI analysis, human review and export distinct so teams can test and govern each stage.
Bring PDFs, scans, images, email-derived material, office documents, spreadsheets and CSV files into a matter-based workspace. Processing exceptions remain visible instead of silently disappearing from the review population.
Tell the AI what should be removed or extracted using the language of the matter. Instructions can focus on personal data, confidential themes, legal advice, settlement material or a named data subject.
RedactAgent maps generated suggestions back to the document or spreadsheet review surface. Suggestions are candidates for review—not an invisible overwrite of the source.
Inspect every proposed removal in context. Reviewers can keep, change or remove marks and route uncertain material through the appropriate legal or privacy decision process.
Create redacted outputs only after review. Production checks, matter history and export evidence help teams account for what was processed, what was withheld and what still requires attention.
Product walkthrough
This synthetic demonstration shows a reviewer entering an AI redaction instruction, running the request and reviewing the resulting marks inside RedactAgent. The workflow is intentionally visible: reviewers see the document, the instruction and the proposed redactions before production.
All public RedactAgent demonstrations use synthetic data. No customer documents are shown in website media.
Use the free redaction QA checklist when assessing the resulting production.
Identify data about the requester, find third-party information, route potentially privileged content and prepare a reviewed disclosure set without losing document-family context.
Stage passages involving legal advice, confidential strategy, settlement discussions or investigation material for qualified human review. RedactAgent supports the workflow but does not replace legal judgement.
Review names, contact information, identifiers, financial references, employment information, health-related material and contextual personal data that simple pattern matching may miss.
Work across searchable PDFs, image-derived pages and OCR-backed scans while keeping proposed redactions tied to the visible document review surface.
Review sensitive content with row, column, sheet and workbook context. This is important when disclosure meaning depends on adjacent cells rather than an isolated string.
Stage review of job information, absence material, disciplinary content, witness references and confidential business information before an authorised reviewer decides what should be withheld.
A useful AI redaction test should include the formats your team actually receives—not only clean, searchable PDFs.
| Content | Review requirement | RedactAgent approach |
|---|---|---|
| PDFs and scans | Maintain page context and cover the intended content at export. | Document review with native or OCR-backed text and page geometry where available. |
| Images | Account for visible text and image-based sensitive content. | Image and OCR-assisted review paths with reviewer validation. |
| Office documents | Preserve readable content and the relationship between imported files. | Controlled import and document-level review within the matter. |
| Spreadsheets and CSV | Keep row, column, sheet and workbook meaning available. | Cell-aware review, spreadsheet targeting and workbook-level context. |
| Email-derived material | Consider message and attachment relationships during review. | Matter-based document handling with family context available to the workflow. |
Redaction mistakes are asymmetric. A false positive may remove useful information and delay disclosure; a false negative may expose personal, privileged or confidential material. Legal AI therefore needs a workflow that supports correction, exception handling and final human accountability.
RedactAgent keeps AI assistance separate from the final production decision. This makes it possible to compare generated suggestions with the source, assess uncertain material and record what changed during review.
RedactAgent does not train or fine-tune AI models on customer documents. Public demonstrations use synthetic data.
Legal review requires controlled access and separation between customer workspaces. Procurement materials document the product's security and privacy approach.
Teams should understand how source documents, derived content, exports and audit records are retained and removed throughout a matter lifecycle.
Evaluate before you rely
Use a representative, versioned test set containing expected easy cases and difficult exceptions. Measure missed sensitive content, incorrect suggestions, placement accuracy, reviewer corrections, unresolved exceptions, export integrity and audit evidence separately for each document type.
A single aggregate “accuracy” percentage can conceal a complete failure on scans, spreadsheets, signatures, privilege or document families. Results should identify the corpus version, application revision, prompts, model configuration, limitations and date of the test.
AI redactions are proposed removals generated from document content and review instructions. RedactAgent stages those proposals for a reviewer to inspect, change or remove before export.
RedactAgent stages AI suggestions for reviewer control. Teams can inspect, change, remove and audit generated redaction marks before production. Final legal and disclosure decisions remain with the reviewing team.
Yes. RedactAgent supports controlled review workflows for PDFs, scans, images, office documents, spreadsheets and CSV files. The review surface depends on the source format and available native or OCR-derived content.
Yes. AI can help stage content involving the data subject, third parties, sensitive personal data and potentially privileged material. Reviewers remain responsible for relevance, exemptions, privilege and the final disclosure.
No. RedactAgent does not train or fine-tune AI models on customer documents.
Use a representative, versioned test set and measure missed sensitive content, incorrect suggestions, placement, reviewer corrections, exceptions, export integrity and audit evidence by document type. Do not rely on one unsupported accuracy percentage.