# Medical Records Redaction Software > Medical records redaction software detects PHI, PII, other-patient identifiers, and privileged content, suggesting redactions for human approval. Canonical page: https://medrecords.ai/product/medical-records-redaction/ --- NewWe now read the actual MRI and CT imaging — not just the radiologist's 1-page report. [See how →](https://medrecords.ai/product/imaging/) [Home](https://medrecords.ai/) ›[Product](https://medrecords.ai/product/) › — PHI Redaction & Production Prep MEDICAL RECORDS REDACTION SOFTWARE ## Produce the file — not the parts that were never yours to produce. Medical records redaction software detects PHI, PII, other-patient identifiers, and privileged content automatically, then suggests each redaction for your approval — per item or per category — before anything is burned in. Approved redactions are applied to a logged, production-ready copy; nothing ships unreviewed, and the original record is never altered. **Suggest-then-approve:** the AI proposes every redaction; a human reviewer approves it, per item or per category, before anything is burned in. Nothing ships unreviewed, and the original record is never altered. [Test a file ](https://medrecords.ai/test-a-file/?src=product-medical-records-redaction) [Book a demo](https://medrecords.ai/demo/) Adams, Timothy — right knee · Case #IME-4812 342 pages · 2 packets Suggested redaction — Category — Status Social Security number PHI ✓ Approved Date of birth PHI Awaiting review Other-patient content — p.140 CO-MINGLED Quarantined Email to defense counsel PRIVILEGED? For counsel Nothing burned in until a reviewer approves — original preserved In action — Case #IME-4812 342 pages scanned for identifiers, the co-mingled p.140 quarantined, and every proposed redaction **held for human approval** before the production copy exists. ### Detection that reads like a reviewer, not a regex. Names, SSNs, dates of birth, MRNs, addresses, and phone numbers are found in context — typed or handwritten, in headers, footers, and margins. 2 categories generic tools miss are first-class here: pages that belong to a different patient, and content that looks privileged. Direct identifiers found in context across every page Other-patient pages and privilege candidates flagged separately Detection categories PHI / PII Names, SSNs, DOBs, MRNs, addresses, phone numbers Other-patient content Co-mingled pages quarantined before they leak Privilege candidates Attorney communications flagged for counsel review Custom terms Case-specific names and terms you add per matter Every detection carries its page location and category — a suggestion, not an action. Redaction profiles same source · separate approvals Opposing counsel Full PHI set + privilege withheld APPROVED Client copy Other-patient content only IN REVIEW Internal working copy Unredacted — access-controlled RESTRICTED ### You approve — per item, per category, per audience. Work through suggestions one by one, or approve a whole category at once when the pattern is clear. Build different redaction profiles from the same suggestions: the set opposing counsel receives is not the set your client receives, and each is approved and logged on its own. Approve or reject individually, or by category in bulk Separate profiles for opposing counsel, client, and internal use ### A production copy with a paper trail. Approved redactions are burned into a separate production set; the original stays intact and access-controlled. Every applied redaction is logged with its page, category, reason code, approver, and timestamp — so "what was withheld, and why" has a documented answer. Original record preserved, never overwritten Exportable log: page, category, reason code, approver Redaction log exportable · per production Page — Category — Reason code — Approved by p.2 PHI · SSN HIPAA-ID Reviewer, K. p.140 CO-MINGLED OTHER-PT Reviewer, K. p.342 PRIVILEGE AC-COMM Counsel, R. Rejected suggestions are logged too — the record shows what was considered, not just what was applied. Review queue · Case #IME-4812 human-gated AI suggests every redaction, cited to its page Human approves per item or per category Then, and only then burned into the production copy The original never altered AI suggests, humans decide ### No automatic redaction ships unreviewed. A missed identifier is a privacy incident; an over-redaction is a discovery fight. Both are human calls. That's why every suggestion is cited to its page, every approval is attributed, and the whole run lands in the audit trail — legally defensible because a person signed off on it. [See Audit Trail & Chain of Custody ](https://medrecords.ai/product/audit-trail-chain-of-custody/) ### From raw record to redacted production. 3 steps, with a human gate in the middle where it belongs. STEP 1 Upload and scan Every page is read and scanned for PHI, PII, other-patient content, and privilege candidates — each detection cited to its page. STEP 2 Review and approve A reviewer works the suggestion queue — approving, rejecting, or escalating to counsel — per item or per category. STEP 3 Produce the logged copy Approved redactions are burned into a production set with its log attached; the original stays intact and access-controlled. ### Who prepares productions with it. Anyone who has to hand a medical file to someone who shouldn't see all of it. [ IME organizations Exam files cleaned of other-patient pages before they reach the evaluator. For IME orgs ](https://medrecords.ai/solutions/medical-evaluators/) [ Law firms Discovery productions redacted, logged, and defensible — privilege flagged for counsel. For law firms ](https://medrecords.ai/solutions/law-firms/) [ TPAs & case management Claim files shared across parties with the right redaction profile per recipient. For TPAs ](https://medrecords.ai/solutions/tpas/) [ Insurance carriers PHI minimization on outbound files, with a log that stands up in an audit. For carriers ](https://medrecords.ai/solutions/insurance-carriers/) FAQ ### Medical records redaction, answered. No. Approved redactions are burned into a separate production copy; the original record is preserved untouched, page for page. You can always go back to the unredacted source, and the platform keeps the 2 clearly separated so the wrong version can't go out by accident. Names, Social Security numbers, dates of birth, medical record numbers, addresses, phone numbers, and similar direct identifiers — plus 2 categories generic tools miss: content belonging to a different patient (co-mingled pages), and communications that look privileged. Every detection is a suggestion with a category and location, not an applied redaction. Yes. A production for opposing counsel, a copy for your client, and an internal working copy can each carry a different approved redaction set — built from the same suggestions, approved separately, and logged separately. One source record, several defensible outputs. Every applied redaction, with its page location, category, reason code, who approved it, and when. Rejected suggestions are logged too. If you're later asked to justify what was withheld and why, the log is the answer — exportable alongside the production set. Pages and passages that look like attorney-client communications or work product are flagged as a distinct category for counsel to review — flagged, never auto-redacted. Privilege calls are legal judgments, so the platform surfaces the candidates and a human makes the call. ### Related capabilities. Redaction sits between organizing the file and producing it — these are its neighbors. [ Bates Numbering & Stamping Sequential numbers and legends on the same production pass. Explore ](https://medrecords.ai/product/bates-numbering-stamping/) [ Co-Mingled Records Detection Wrong-patient pages quarantined before analysis or production. Explore ](https://medrecords.ai/product/co-mingled-records-detection/) [ Audit Trail & Chain of Custody An immutable log of every access, edit, and export on the file. Explore ](https://medrecords.ai/product/audit-trail-chain-of-custody/) [ Document Sorting & Packet Builder Rule-based sorting and packet assembly ahead of production. Explore ](https://medrecords.ai/product/document-sorting-packet-builder/) ### Produce a redacted set you can defend. Upload a file and see the suggested redactions, the review queue, and the log — before anything is burned in. Handled under our BAA; never used to train a model. [Test a file ](https://medrecords.ai/test-a-file/?src=product-medical-records-redaction) [Book a demo](https://medrecords.ai/demo/) [All capabilities ](https://medrecords.ai/product/)