Appearance
Healthcare
GhostAI provides specialized protection for healthcare organizations handling Protected Health Information (PHI).
Overview
Healthcare organizations face unique challenges when adopting AI:
- PHI Protection: Patient data is highly regulated
- HIPAA Compliance: Strict requirements for data handling
- Clinical Workflows: Need AI assistance without risking privacy
- Audit Requirements: Detailed logging for compliance
GhostAI addresses all of these with healthcare-specific features.
Healthcare Shield Add-On
The Healthcare Shield add-on provides HIPAA-grade PHI detection, BAA coverage, clinical NER models, and healthcare-specific compliance reporting. Available on Personal and above at $15/user/mo.
Quick Start for Healthcare
1. Select Healthcare Vertical
In Settings → Industry Vertical, select Healthcare.
This enables:
- Enhanced PHI detection (18 HIPAA identifiers)
- Lower detection thresholds for medical data
- HIPAA-specific compliance tagging
- Healthcare-focused audit reports
2. Enable PHI-Specific Detection
The healthcare vertical enables detection of:
| PHI Type | Examples |
|---|---|
| Patient Names | John Doe, Dr. Smith |
| MRN | Medical record numbers |
| Dates | DOB, admission, discharge |
| SSN | Social Security Numbers |
| Phone/Fax | Patient contact info |
| Patient email addresses | |
| Address | Street, city, ZIP |
| Account Numbers | Insurance IDs |
| Certificate Numbers | License numbers |
| Device IDs | Pacemaker serials |
| URLs | Patient portal links |
| IP Addresses | EHR access logs |
| Biometric | Fingerprints (descriptions) |
| Photos | Referenced images |
| Age (over 89) | Specific ages |
| Geographic Data | Detailed locations |
Public Figure Reconciliation
Names of well-known physicians, researchers, and public health figures (e.g., "Dr. Fauci", "Jonas Salk") are recognized as public figures and not flagged as PHI. This reduces false positives in clinical and research discussions while still protecting actual patient data. Co-occurrence and multi-detector safety filters ensure that patient names are never mistakenly reconciled as public figures, even when a patient shares a name with a well-known individual.
Public Knowledge Reconciliation
Healthcare teams can also reconcile public LOCATION, ORGANIZATION, and EVENT entities in research and policy discussions. ORGANIZATION reconciliation uses a curated list (and optional tiers), not a registry of public companies. HIPAA-sensitive context remains protected: FACILITY_NAME and private clinical date context are still redacted by safety filters.
3. Review Compliance Settings
In Settings → Compliance:
- [ ] Enable HIPAA tagging
- [ ] Enable HITECH tagging
- [ ] Set retention for 6 years (HIPAA minimum)
- [ ] Enable audit logging
- [ ] Configure access controls
Use Cases
- PDF Generation for HIPAA-compliant patient briefings, clinical summaries, and medical reports — with rich tables (lab results, vitals), Mermaid diagrams (care workflows, referral paths), and embedded images for clinical documentation. Zero-trust secure PDF mode protects PHI in generated reports: data tool results are tokenized before reaching the LLM; PDF content is detokenized server-side at render time with full audit trail.
Clinical Documentation
Scenario: Physician needs AI help drafting clinical notes.
Before GhostAI:
Patient John Smith (MRN: 12345678) presents with chest pain.
DOB: 01/15/1965, SSN: 123-45-6789.
Contact: (555) 123-4567, john.smith@email.comWhat AI Sees:
Patient [PERSON_a1b2] (MRN: [MRN_c3d4]) presents with chest pain.
DOB: [DATE_e5f6], SSN: [SSN_g7h8].
Contact: [PHONE_i9j0], [EMAIL_k1l2]AI Response (detokenized):
Based on the patient John Smith's presentation with chest pain,
I recommend the following workup...Patient Referrals
Scenario: Coordinator sending referral information.
GhostAI protects:
- Patient identifiers
- Insurance information
- Medical history details
- Contact information
While allowing the AI to help with:
- Referral letter drafting
- Appointment scheduling suggestions
- Follow-up care recommendations
Medical Coding
Scenario: Coder needs help with ICD-10 codes.
GhostAI allows:
- Medical terminology (not PHI)
- Procedure descriptions (anonymized)
- Diagnosis discussion (without patient ID)
Research & Quality
Scenario: Quality team analyzing patient outcomes.
GhostAI enables:
- Statistical discussions
- De-identified data analysis
- Best practice queries
HIPAA Compliance Details
18 HIPAA Identifiers
GhostAI detects and protects all 18 HIPAA identifiers:
- Names
- Geographic data (smaller than state)
- Dates (except year)
- Phone numbers
- Fax numbers
- Email addresses
- SSN
- Medical record numbers
- Health plan beneficiary numbers
- Account numbers
- Certificate/license numbers
- Vehicle identifiers
- Device identifiers
- URLs
- IP addresses
- Biometric identifiers
- Full-face photos
- Any other unique identifier
Minimum Necessary Standard
GhostAI helps enforce the minimum necessary standard by:
- Detecting what PHI is being shared
- Prompting users to consider if all PHI is needed
- Offering selective tokenization
Audit Controls
HIPAA-compliant audit logging includes:
| Logged | Details |
|---|---|
| Who | User ID, role |
| What | PHI accessed/shared |
| When | Timestamp |
| Where | IP, device |
| Why | Session context |
BAA Availability
GhostAI offers Business Associate Agreements for:
- Team tier and above (with Healthcare Shield)
- Covered entity protection
- Subcontractor compliance
Medical File Support
DICOM Processing
GhostAI processes DICOM medical images:
- Patient Tags: Name, ID, DOB extracted and protected
- Study Info: Dates and descriptions scanned
- Burned-in Text: OCR for image overlays
- Metadata Stripping: Remove identifying metadata
HL7 Message Parsing
GhostAI parses HL7 messages:
- PID Segments: Patient identification
- NK1 Segments: Next of kin
- PV1 Segments: Visit information
- OBX Segments: Observations
Healthcare-Specific Reports
HIPAA Compliance Report
Generate monthly/quarterly reports:
- Total PHI items protected
- PHI types distribution
- User access patterns
- Potential violations
- Remediation recommendations
Board Report for Healthcare
Executive summary including:
- Cost avoidance (HIPAA breach cost: ~$150/record)
- Risk reduction metrics
- Compliance score
- Incident trends
Best Practices
For Clinicians
- Review highlights: Check what GhostAI flags before sending
- Use templates: Pre-approved prompts reduce risk
- Minimal PHI: Only include necessary identifiers
- Verify responses: Ensure AI didn't hallucinate PHI
For IT/Compliance
- Enable all PHI detection: Don't disable healthcare-specific types
- Set appropriate thresholds: Lower for sensitive data
- Regular audits: Review access logs monthly
- Training: Educate staff on GhostAI use
For Administration
- Role-based access: Limit who can view audit logs
- Data residency: Ensure data stays in appropriate region
- Retention policies: Align with HIPAA requirements
- Incident response: Define procedures for potential breaches
Memory & HIPAA Compliance
AI Memory in healthcare deployments is subject to additional HIPAA safeguards:
- PHI in Memories: All memories pass through the full PHI detection pipeline before storage. Patient names, MRNs, dates, and other HIPAA identifiers are tokenized in memory content just as they are in chat messages.
- Retention Policies: Organization admins can set memory retention limits aligned with HIPAA minimum retention (6 years). Expired memories are automatically purged with audit trail.
- Access Controls: Memory scopes respect HIPAA role-based access. Organization and group memories are restricted to authorized users within the covered entity.
- Audit Trail: Every memory creation, access, update, and deletion is logged with who/what/when/where/why metadata for HIPAA audit requirements.
- BAA Coverage: Memories stored under the Healthcare Shield add-on are covered by the Business Associate Agreement, same as chat data.
- Minimum Necessary: Users are prompted to review memory content before sharing to organization or group scopes, ensuring only necessary information is retained.
- Memory Interleave (HIPAA): Memory Interleave respects HIPAA requirements: no PHI in tool queries, and all
recall_memoryinvocations are audit-logged withsession_idandquery_hash.
Company Intelligence Profile for Healthcare
The Company Intelligence Profile amplifies healthcare-specific detection:
- Pre-seed patient ID formats: Register your organization's MRN format, facility codes, and internal patient identifiers as sensitive terms so Ghost AI catches them even when standard PHI patterns miss
- Internal drug codenames: Add pre-approval drug names, compound identifiers, and clinical trial codes that would otherwise bypass generic detection
- Facility and provider names: Mark your facility names and provider identifiers as sensitive context so the AI understands when they appear alongside patient data
- Safe terms: Register publicly known facility names, department abbreviations, and standard medical terminology that frequently trigger false positives
- Auto-Research: One-click profile population from public sources identifies your organization's specializations and suggests relevant terms
Fine-tuning detection for Healthcare
To improve detection for healthcare-specific entities (e.g. MRN formats, facility codes, internal identifiers), you can fine-tune the PII model on your labeled data. See PII Fine-Tuning and the True ML Fine-Tuning whitepaper for data format and per-vertical examples.