Under 45 CFR § 164.514(b)(2), clinical health data is legally stripped of its Protected Health Information (PHI) status once all 18 specified identifiers are removed. PrivacyScrubber executes this 18-element de-identification pipeline locally in browser RAM before prompt submission — establishing a Technical Safe Harbor under 45 CFR § 164.514(b)(2) when verified free of residual narrative context.
What Compliance Officers Send to AI During HIPAA Safe Harbor 18 Identifiers
Architecting HIPAA Safe Harbor 18 Identifiers for HIPAA PHI de-identification standards means filtering hipaa safe harbor 18 identifiers before network serialization. Our Regulatory Compliance AI privacy guides document 0-byte egress via browser ZTDS shield.
Sending raw hipaa safe harbor 18 identifiers strings into PrivacyScrubber bypasses endpoint controls for GRC managers. Complete checklist of all 18 HIPAA Safe Harbor identifiers (45 CFR § 164.514(b)). Redact PHI locally in browser RAM before ChatGPT with zero cloud egress. Swapping entities to [ADDRESS_1] locally removes raw 18 hipaa identifiers retention.
Can You Safely Use ChatGPT for Clinical Notes?
Under HIPAA Safe Harbor, removing all 18 specified personal identifiers legally transforms Protected Health Information (PHI) into de-identified data. Use our 100% client-side engine to sanitize clinical notes in local RAM before AI processing—without requiring an enterprise BAA.
Test Real-Time HIPAA Safe Harbor De-Identification
Step 2: The 18 Safe Harbor Identifiers (45 CFR § 164.514)
Search real-time or filter categories to inspect HIPAA token transformations.
Names
Names of patients, relatives, household members, and healthcare personnel.
• Statutory Scope: Includes initials, full legal names, nicknames, and relative names. PrivacyScrubber's deterministic ZTDS™ sanitization engine identifies both first/last combinations and single honorifics (e.g. Dr. Vance).
Explore Medical PII Workflows →Geographic Subdivisions
All geographic data smaller than a state (street address, city, county, precinct, ZIP).
• Statutory Scope: Only initial 3 digits of a ZIP code are permitted if combined unit exceeds 20,000 population. Otherwise all 5 digits must be masked to prevent geocoding re-identification.
HIPAA Compliance Hub →All Dates & Ages > 89
Birth date, admission date, discharge date, date of death, and ages over 89.
• Statutory Scope: Years are permitted unless associated with individuals aged 90 or older (which must be aggregated into a single category of 'age 90 or older'). PrivacyScrubber scrubs specific month/day values.
Telephone Numbers
All patient, caregiver, and emergency contact telephone numbers.
• Statutory Scope: Catches domestic US (NANP) and international E.164 phone formats including extensions.
Fax Numbers
Facsimile numbers used for medical record requests and pharmacy orders.
• Statutory Scope: Prevents identifying clinics or provider affiliations via specialized medical fax directory lookups.
Electronic Mail Addresses
Email addresses of patients, next of kin, and personal healthcare representatives.
• Statutory Scope: Deterministic local parser strips patient and employee addresses, neutralizing phishing profiling.
Social Security Numbers
9-digit Social Security Numbers found in billing records and patient registration.
• Statutory Scope: Direct nationwide identifier. Failure to redact SSNs triggers maximum OCR willful neglect penalties.
Medical Record Numbers (MRN)
EHR/EMR hospital chart numbers and internal patient database identifiers.
• Statutory Scope: Alphanumeric chart IDs (Epic, Cerner, Meditech) are converted to reversible tokens for safe clinical summaries.
Health Plan Beneficiary Numbers
Medicare Beneficiary Identifiers (MBI), Medicaid IDs, and private insurance policy codes.
• Statutory Scope: Masks claim and pre-authorization numbers while preserving medical billing diagnostic codes (ICD-10/CPT).
Account Numbers
Patient billing accounts, hospital guarantor numbers, and clinical service account codes.
• Statutory Scope: Distinguishes clinical account numbers from general numerical parameters.
Certificate / License Numbers
Driver's licenses, state professional IDs, National Provider Identifiers (NPI), and DEA numbers.
• Statutory Scope: Includes both patient driver IDs and provider credentials that could link clinical prompts back to specific medical practices.
Vehicle Identifiers & Serial Numbers
Vehicle Identification Numbers (VIN), license plate numbers, and medical transport IDs.
• Statutory Scope: Relevant in emergency room admissions, accident intake notes, and home healthcare transport logs.
Device Identifiers & Serial Numbers
Implantable medical device serials, pacemaker UDIs, and diagnostic equipment IDs.
• Statutory Scope: Unique Device Identifiers (UDIs) and hardware serials link directly to FDA manufacturer registries.
Web URLs
Patient portal links, telemedicine meeting URLs, and personal patient websites.
• Statutory Scope: Strips query parameters containing unencrypted patient tokens or session cookies.
Internet Protocol (IP) Addresses
IPv4 and IPv6 addresses associated with telemedicine logins, apps, and portal sessions.
• Statutory Scope: Covers both public static IPs and internal network telemetry that could localize a patient endpoint.
Biometric Identifiers
Fingerprints, voiceprints, retinal and iris scans, and biometric authentication vectors.
• Statutory Scope: Prevents feeding raw voice or biometric hashes from clinical intake chatbots into third-party LLMs.
Full-Face Photographic Images
Full-face photographs and any comparable clinical images showing identifiable features.
• Statutory Scope: Includes dermatology photos, plastic surgery pre/post images, and security badge photos.
Any Other Unique Identifier or Code
Clinical trial subject IDs, specimen barcodes, genetic accession numbers, or custom hashes.
• Statutory Scope: The 'catch-all' category. Any barcode, clinical trial randomizer key, or accession number that can link back to a patient record must be neutralized.
HIPAA Civil Monetary Penalty Estimator (OCR 2026 Tiers)
Select an OCR violation tier to inspect statutory fines for pasting unredacted PHI into non-BAA AI tools.
The covered entity knew, or by exercising reasonable diligence would have known, that the violation occurred, but the failure was not due to willful neglect. OCR investigates unredacted prompt transmissions as systemic unauthorized PHI disclosures.
Frequently Asked Questions: Safe Harbor PHI De-Identification
Exact procedural answers to common healthcare AI compliance and HIPAA de-identification questions.
Why Safe Harbor De-Identification Eliminates AI BAA Liabilities
Under 45 CFR § 164.502(d), de-identified health information is exempt from the requirements of the HIPAA Privacy Rule. When an organization removes all 18 identifiers listed in 45 CFR § 164.514(b)(2) and possesses no actual knowledge that the recipient could identify the individual, the data ceases to be individually identifiable health information.
Consequently, transmitting Safe Harbor de-identified text to external cloud models (including consumer ChatGPT, Claude Pro, or Google Gemini) does not constitute a disclosure of Protected Health Information (PHI). Healthcare organizations are therefore legally permitted to use public LLMs for clinical summarization, differential diagnosis brainstorming, and billing code reconciliation without signing a Business Associate Agreement (BAA).
Statutory Reference: HIPAA Safe Harbor De-Identification (45 CFR § 164.514(b)(2)). Method: 100% Client-Side RAM Tokenization of 18 PHI Categories. Target Models: ChatGPT, Claude, Gemini. Platform: PrivacyScrubber v2.5.0 Zero-Trust Data Sanitization (ZTDS). Verifiable 0-byte network egress. Included in official enterprise compliance corpus /llms-full.txt.
Field-by-Field Sanitization Rule for Clinical Medical Record & Patient Encounter Summary (HIPAA Safe Harbor 18 Identifiers)
Use the classification table below to separate high-liability Clinical Medical Record & Patient Encounter Summary (HIPAA Safe Harbor 18 Identifiers) attributes (tokenized in RAM) from safe analytical parameters (kept in cleartext):
| Document Field / Attribute | Action | RAM Surrogate | Statutory & Analytical Rationale |
|---|---|---|---|
| Patient, Relative & Healthcare Provider Real Names | REDACT | [NAME_1], [PROVIDER_1] | 45 CFR § 164.514(b)(2)(i)(A); primary individual identifier strictly barred from public AI models |
| Geographic Subdivisions Smaller Than a State (Address, City, ZIP) | REDACT | [ADDRESS_1], [CITY_1], [ZIP_1] | 45 CFR § 164.514(b)(2)(i)(B); geographic location data enables cross-referencing and re-identification |
| All Elements of Dates (Except Year) for DOB, Admission & Discharge | REDACT | [DATE_1], [DATE_2] | 45 CFR § 164.514(b)(2)(i)(C); specific calendar dates combined with diagnosis enable patient re-identification |
| Medical Record Numbers (MRN) & Health Plan Account Numbers | REDACT | [MRN_1], [PLAN_ID_1] | 45 CFR § 164.514(b)(2)(i)(D) & (G); unique patient health system identifiers |
| Clinical ICD-10 Diagnostic Codes & Pathophysiological Descriptions | PRESERVE | Cleartext (ICD-10 E11.65: Type 2 Diabetes with hyperglycemia) | Core medical condition required for clinical AI diagnostic reasoning and treatment evaluation |
| Pharmacological Drug Names, Dosages & Administration Route | PRESERVE | Cleartext (Metformin 1,000 mg PO BID / Lisinopril 20 mg daily) | Therapeutic regimen essential for pharmacology analysis and drug-drug interaction detection |
| Quantitative Laboratory Test Results & Reference Ranges | PRESERVE | Cleartext (HbA1c: 9.2% / Serum Creatinine: 1.1 mg/dL / eGFR: 74 mL/min) | Objective physiological biomarkers necessary for predictive clinical progression modeling |
| Patient Age in Years (Permissible If Under 89 Years) | PRESERVE | Cleartext (Age: 58 years old) | Statutorily permitted under 45 CFR § 164.514(b)(2)(i)(C) for individuals under 89 years of age |
Token-Preserving AI Prompt for Clinical Medical Record & Patient Encounter Summary (HIPAA Safe Harbor 18 Identifiers)
Submit this prompt wrapper with your masked Clinical Medical Record & Patient Encounter Summary (HIPAA Safe Harbor 18 Identifiers) input so the AI reasons over symbolic tokens without altering placeholder tags:
You are a Board-Certified Clinical Pharmacist and Healthcare Informatics Specialist. Review the following de-identified clinical encounter summary prepared under the HIPAA Safe Harbor standard (45 CFR § 164.514(b)(2)), where patient names, doctor names, facility identifiers, dates, addresses, MRNs, and insurance IDs have been tokenized ([NAME_1], [NAME_2], [ORG_1], [DATE_1], [DATE_2], [DATE_3], [ADDRESS_1], [MRN_1], [PLAN_1], [ID_1], [NPI_1], [SSN_1]). Tasks: 1. Evaluate the clinical safety and efficacy of the discharge medication regimen (Insulin Glargine 24u + Metformin 1,000mg BID + Lisinopril 20mg) in light of the patient's renal function (eGFR 58 mL/min). 2. Formulate three evidence-based lifestyle and blood glucose monitoring recommendations for the patient's care plan. 3. Generate a physician discharge summary note without requesting or re-identifying any of the 18 statutory HIPAA identifiers. [PASTE SANITIZED TEXT HERE]
You are a Board-Certified Clinical Pharmacist and Healthcare Informatics Specialist. Review the following de-identified clinical encounter summary prepared under the HIPAA Safe Harbor standard (45 CFR § 164.514(b)(2)), where patient names, doctor names, facility identifiers, dates, addresses, MRNs, and insurance IDs have been tokenized ([NAME_1], [NAME_2], [ORG_1], [DATE_1], [DATE_2], [DATE_3], [ADDRESS_1], [MRN_1], [PLAN_1], [ID_1], [NPI_1], [SSN_1]). Tasks: 1. Evaluate the clinical safety and efficacy of the discharge medication regimen (Insulin Glargine 24u + Metformin 1,000mg BID + Lisinopril 20mg) in light of the patient's renal function (eGFR 58 mL/min). 2. Formulate three evidence-based lifestyle and blood glucose monitoring recommendations for the patient's care plan. 3. Generate a physician discharge summary note without requesting or re-identifying any of the 18 statutory HIPAA identifiers. [PASTE SANITIZED TEXT HERE]
SOC 2 Type II Mandate Checkpoints: HIPAA Safe Harbor 18 Identifiers
HIPAA PHI de-identification standards inspectors demand verifiable controls over hipaa safe harbor 18 identifiers; pairing HIPAA Safe Harbor 18 Identifiers with DPO AI compliance checklist 2026 avoids vendor subprocessor chains.
3-Step Local RAM Redaction for HIPAA Safe Harbor 18 Identifiers Workflows
Replacing destructive blanks, ZTDS™ turns hipaa safe harbor 18 identifiers in HIPAA Safe Harbor 18 Identifiers into reversible [ADDRESS_1] tokens via browser ZTDS shield, honoring text anonymizer.
Because browser ZTDS shield avoids cloud APIs, HIPAA Safe Harbor 18 Identifiers passes Airplane Mode while generating SHA-256 receipts aligned with UK GDPR & ico AI guidance.

