AI assistance off
The planned security checks do not require a model. This website makes no Shield model calls and does not process security evidence.
PRIVACY & PRIVATE AI
Protect patient information.
Control where it goes.
Your patients trust you with personal information, clinical photographs and private conversations. Shield’s privacy design keeps purpose, access and processing destinations explicit, so adding security does not mean handing over unrestricted patient data.
Business inquiries welcome · Shield services are in development · View current availability
01 / WHAT THE PRIVACY VIEW WOULD SHOW
Your information.
Clear answers about its use.
- Information and purpose
- Which security records are needed and why. Patient charts, photos, messages and credentials are not automatically approved for collection or AI use.
- Processing location
- The actual service, model provider and region involved—including supporting search, storage, logs and caches. These are not selected for a live practice deployment yet.
- AI permissions
- Whether assistance is disabled, private-only or separately approved for an external provider. An alternative provider cannot be used simply because the first request failed.
- Retention and deletion
- How long each kind of information may be kept and how deletion or a required hold would work. Live terms must be agreed; no blanket retention period is promised here.
- Support access
- Who may assist, what they may access and when that permission expires. Commercial enrollment does not grant unrestricted support access.
- Verification
- What has actually been tested. A written policy or a model called “private” does not prove that every processing route follows the rules.
02 / AI OPTIONS IN THE PRODUCT DESIGN
Private by agreement.
Checked in practice.
Private-only assistance
Only an explicitly approved private deployment could receive the agreed information. Its location, terms, access and retention still need verification.
Approved external assistance
A named external provider could receive only the separately approved information for the agreed purpose. External AI is not active in this product preview.
03 / AI ASSISTS. PEOPLE AUTHORIZE.
Useful help.
Without handing over authority.
- 01
Use only permitted information
Limit a request to the facts needed for the approved task. Do not send credentials or unrestricted clinical content.
- 02
Keep the evidence attached
A suggested explanation must distinguish what was observed, what is inferred and what is unknown.
- 03
Check permission again
Access and data-use approvals must still be valid when information is sent and when a result is used.
- 04
Keep consequential decisions with people
AI cannot grant access, approve its own correction, decide legal liability or replace clinical judgment.
KNOWLEDGE & CONSTRAINED AI
Useful context.
No authority invented by a model.
Explore the knowledge and AI controls
These are implemented offline or synthetic references. Live AI transport, production request storage and permitted deployment evidence remain separate work.
- Context without invented certainty
- The reference vocabulary links accounts, assets, data categories, purposes, patient-choice records, evidence and current authority. A matching name does not prove two records represent the same person or resource.
- Source-linked knowledge
- Authorized retrieval is restricted by practice, person, purpose, source, sensitivity and time. The offline reference keeps observations, interpretations and unknowns separate; it is not a complete legal database or free-text research engine.
- Permission before model selection
- The router first checks the exact task, permitted information, destination and current approvals. Only eligible routes may be compared for quality, latency and cost. The current ranking estimates are synthetic, not measured model performance.
- Bounded case assistance
- The synthetic task can organize permitted facts and follow-up checks with their sources and uncertainties. Conflicting evidence remains visible. Output validation does not establish that a real model is useful or correct.
- Current authority through the whole request
- The local persisted request reference checks permission before dispatch and when accepting a result. Late, replayed or stale-authority results are rejected; ambiguous sends require reconciliation, not an automatic retry.
PATIENT INFORMATION & PERMITTED USE
Follow the purpose.
Across every destination.
- Sensitive information stays sensitive
- Clinical images, scans, derived face/body information, messages, identity records and security metadata may require different handling. Removing names or using opaque references does not prove information is anonymous.
- Purpose stays explicit
- Clinical care, patient communication, marketing, security investigation and model training are separate uses. Buying Shield or approving a security scope cannot substitute for those permissions.
- All processing routes count
- Model inference, embeddings, logs, traces, caches, tools, support access and evidence storage need explicit destinations and retention rules. A private answer model does not make a cloud-connected pipeline fully local.
- Failure cannot widen permission
- A private model outage cannot silently switch a request to a public provider. No eligible route means no model call; deterministic checks remain independent of AI.
PRIVACY & LEGAL REVIEW DESIGN
Preserve the evidence.
Respect the review boundary.
The right questions for the right adviser
The legal/privacy intelligence specification organizes candidate review issues, missing facts and source dates. The responsible privacy or legal reviewer determines what applies; the system does not make a final breach or notification decision.
AI and analytics transfers
Review the information category, purpose and destination together. Record missing permission evidence without treating it as proof of an unlawful transfer.
Clinical media and derived data
Distinguish a photo, clinical image, geometry and identity-related use. Do not automatically label every photograph regulated biometric information.
YOUR NEXT STEP
Identify the weak points.
Plan the next step.
Tell us which systems your practice uses and what you want to protect. We’ll discuss the relevant scope, readiness and responsibilities before any work.