Trigger: AI may access patient, clinical, or wellbeing data.
Health AI Evidence Passport
Health AI needs evidence
before it reaches patients.
Local evidence, raw-export flags, and access decisions for patient-facing AI — without clinical data leaving your perimeter.
Data Boundary
Raw health data
stays inside perimeter
Signal Receipts
travel only
Sensitive context
Why health AI is a high-sensitivity context.
AI tools operating near patient data, clinical decisions, or health records handle some of the most protected personal data in existence. Even indirect access — a chatbot that processes care history, a note assistant that handles clinical text, or a recommendation engine with wellness data — creates significant risk if evidence is absent.
Data categories in scope
People affected
Risk scenarios
What typically goes wrong.
Specific failure modes seen in this sensitive context — without structured evidence.
A patient companion chatbot sends conversation context to an external LLM API.
No evidence of whether raw patient narrative leaves the perimeter. No DPA confirmed. GDPR Art. 9 health data exposed without legal basis.
A clinical note assistant processes consultation transcripts with a general-purpose AI.
No raw-export flag visible to DPO. No confirmation that transcripts are not retained or used for training. No DPIA completed.
A wellbeing app integrates an AI layer with no subprocessor list.
Third-party model providers unknown to the commissioning organisation. Signal Receipts absent. No renewal monitor.
An imaging AI prototype is piloted without evidence of data-boundary configuration.
Raw DICOM images potentially accessible outside the clinical network. No evidence of local-only operation.
Vendor claims NHS/GDPR alignment — no structured evidence provided.
Procurement cannot distinguish claim from evidence. DPO cannot verify. Access decision made on trust, not evidence.
Scope
What needs a Passport.
Stakeholder workflow
From trigger to access decision.
Trigger
AI system in scope
Evidence Request
Passport initiated
Review
DPO · CISO · Specialist
Decision
Access condition set
Monitor
Tide sweeps · Renewal
Trigger
AI system in scope
Evidence Request
Passport initiated
Review
DPO · CISO · Specialist
Decision
Access condition set
Monitor
Tide sweeps · Renewal
DPO
“An AI tool may process Art. 9 health data or special category data.”
Request Evidence Passport with legal basis, DPA, and DPIA status before sign-off.
CISO
“A health AI vendor has not confirmed data residency or encryption posture.”
Review security section of the Passport. Require raw-export flag = off.
Clinical Owner
“A clinical note assistant is being trialled without formal evidence review.”
Require Passport before clinical staff use the tool on patient data.
Access decisions
Context Visa conditions.
The access decisions that apply in this sensitive context — and the evidence conditions that produce them.
- All inference runs on-premises or within clinical network
- No raw patient data leaves the perimeter
- Signal Receipts only leave for Passport record
- Anonymised or pseudonymised data only
- DPA signed and confirmed
- DPIA completed and accepted by DPO
- Renewal in 12 months
- System operates within clinical decision support
- Human clinician must verify AI outputs before patient impact
- Audit trail required for each interaction
- Evidence incomplete — DPIA or DPA missing
- Subprocessor list not confirmed
- Raw export flag not set
- Raw health data confirmed as leaving clinical perimeter without legal basis
- No DPA
- GDPR Art. 9 basis absent
Measurement
Evidence families we can structure.
The measurable evidence categories relevant to this context and the evidence signals they produce.
Privacy & Legal Basis
GDPR Art. 6 and Art. 9 legal basis, DPA status, subprocessors, data residency, and raw-export flag.
Data Boundary
Evidence that raw patient data does not leave the clinical perimeter. Signal Receipts travel; raw assets do not.
Security Posture
Encryption at rest and in transit, access controls, audit logging, and vendor certifications.
Human Oversight
Confirmation that clinical decisions are reviewed by qualified staff before patient impact.
Vendor Evidence
Subprocessor list, data residency confirmation, training-data policy, and model provider stack.
DPIA Status
Whether a Data Protection Impact Assessment has been completed, accepted, and reviewed by the DPO.
Honest scope
What remains not assessable.
AffectLog does not overclaim. These items require external expertise, regulatory process, or long-term study.
Clinical validity or diagnostic accuracy
AffectLog measures technical and operational evidence — not clinical effectiveness. Diagnostic accuracy requires clinical study, not AI governance tooling.
Instead: Reference external clinical evaluation, MHRA registration, or CE marking for medical devices.
Medical device regulatory compliance (MDR/IVDR)
Medical device classification and conformity assessment is a regulatory process requiring a notified body, not an evidence platform.
Instead: Commission a regulatory consultant or notified-body review for AI systems meeting the MDR definition.
Whether an AI tool is safe for clinical use
Safety for clinical deployment requires validated efficacy evidence and clinical governance — outside AffectLog scope.
Instead: Engage clinical governance and patient safety teams before deployment.
Example
Sample Passport for this context.
Patient Companion Chatbot
Patient-facing wellbeing assistant · NHS / Healthcare
Access conditions
What we will not overclaim
AffectLog provides technical and operational evidence for health AI access decisions. We do not claim clinical validity, medical-device approval, or regulatory certification. We show what evidence exists, what remains not assessable, and what review conditions apply.
Common questions
Questions this context raises.
“Our health AI vendor is already GDPR compliant — we have their DPA.”
A DPA is a legal contract. AffectLog structures the technical evidence that sits behind it: which patient data categories are processed, whether raw data leaves your perimeter, which subprocessors are involved, and what the model provider does with conversation data.
“We cannot send patient data to any external tool to assess it.”
You do not need to. The local runner executes diagnostics inside your clinical network. Only signed Signal Receipts leave — never raw records, prompts, or clinical text.
“We are too early in the pilot to need formal evidence.”
Pilots are exactly when evidence matters most. A pilot with no Passport and no data boundary is a live system. Structured evidence at pilot stage prevents a larger incident later.
Get started
Keep health AI inside your perimeter
until the evidence supports travel.
Estimate the evidence scope for your health AI portfolio. Identify which systems need local-only conditions, which need DPIAs, and which vendors still owe you a structured Passport.
AffectLog provides technical and operational evidence to support access decisions. Not clinical validation, regulatory certification, or legal advice.