Patient data and LLMs
don't mix unless
you govern them.
Hospitals, clinics, and health-tech companies deploy AI agents for triage, documentation, and clinical decision support. TapPass ensures patient data never reaches the model unprotected.
Why healthcare AI needs a governance layer
Patient data is GDPR Art. 9 special category data, the highest protection level. AI agents in healthcare operate in a regulatory minefield.
Patient PII in LLM prompts
Your documentation agent sends patient names, diagnoses, and treatment plans to an external LLM. GDPR Art. 9 requires explicit consent and strict processing safeguards.
No auditability for clinical AI
The EU AI Act classifies medical AI as high-risk. You need full logging of every AI decision: what data went in, what came out.
Prompt injection in clinical tools
A patient enters malicious text in a symptom field. Your triage agent follows the injected instruction instead of analysing symptoms.
Cross-patient data leakage
AI agents with shared context windows leak patient A's history into patient B's session. Without session isolation, no data separation.
Health data leaving the EU
Routing patient data to a US-hosted LLM violates GDPR transfer rules. Health data carries the highest protection level.
Research–clinical contamination
AI agents processing both clinical and research data risk cross-contamination. De-identified data becomes re-identifiable in LLM context.
Runtime governance for healthcare AI
Detect, block, and audit every patient data touchpoint, every LLM call, every tool invocation.
Patient data detection
Detects names, MRNs, ICD codes, medications, and dozens of obfuscation techniques before data reaches the LLM.
- GDPR Art. 9 special category awareness
- PII tokenisation so the model never sees real data
- Configurable: block, redact, or tokenise
EU AI Act compliance logging
Hash-chained audit trail captures every AI decision. Classification, detections, data categories, timing.
- Art. 12 record-keeping, tamper-evident
- Art. 14 human oversight with approval gates
- SIEM export for hospital SOC
Clinical safety guardrails
Healthcare guardrail rules prevent AI agents from making unsupported clinical claims or overriding human judgment.
- Human approval for clinical decisions
- Output scanning for medical claims
- Behavioural pacts enforce agent scope
Session isolation
Session-scoped taint tracking ensures patient A's data never leaks into patient B's session.
- Per-patient session boundaries
- Taint propagation tracking
- Memory poisoning detection
EU-only data routing
Force all patient data to EU-hosted providers. Self-hosted deployment for maximum control.
- Classification-based provider routing
- Self-hosted deployment option
- Zero data leaves your network
GDPR erasure (Art. 17)
Cryptographic erasure with tombstones. Patient's right to erasure while preserving audit chain integrity.
- Signed erasure receipts
- Hash chain recomputation
- Configurable retention policies
Patient data deserves more than a privacy policy.
Scan, classify, and audit every AI interaction with patient data. EU AI Act ready.