AI Agent Security Review
Healthcare AI Deployment Review
Review patient-data flows, tool permissions and deployment controls before launch or a customer security review.
Who it is for
For digital-health vendors, healthcare organizations and teams introducing AI agents into workflows that involve sensitive health information. It is a defined engagement within our AI Agent Security Review.
Best fit: before you launch an agent that handles patient data, give it new tools or access, or answer a healthcare customer's security review.
The review shows how health information moves through the system, what the agent is allowed to do with it, and what to fix first.
What the review can examine
We agree the scope in writing before any testing starts. Depending on your system, it can cover:
- Patient-data flows across the interface, model endpoints, retrieval stores, connectors, tool execution, logs and external services.
- Agent identity, access permissions and the authority each tool gives the agent.
- Selected prompt-injection, unauthorized-action and data-exposure scenarios, run in an agreed environment.
- Vendor evidence for the product tier and deployment configuration you actually use.
- Monitoring, incident visibility, and the changes that should trigger a new review.
Tests run only within the authorized scope, on the environment and data agreed beforehand. Anything we could not examine is recorded as a limitation.
What you receive
- An annotated data-flow diagram showing where health information goes.
- A permissions matrix: which identities and tools can read, change or act, and on what.
- Test evidence for each scenario run.
- Prioritized findings, each with its supporting evidence.
- Evidence gaps and the limitations of the review.
- An agreed remediation and retest plan. If Medoya helps implement the fixes, the retest is remediation verification, not an independent assessment.
How it works
Step 1
Book a free call
30 minutes about the system and how it is used. Bring your deadline and any customer security questions. We'll discuss access, your team's involvement, evidence handling and report-sharing requirements. It is a fit discussion, not an assessment.
Step 2
Agree the scope
Systems, access, authorized tests, exclusions, deliverables and fees, in writing before any work starts.
Step 3
Assess and test
We examine the agreed material, run the authorized checks and record what remains unverified.
Step 4
Explain the findings
Evidence, limitations and recommendations, so you can decide what to fix and whether more work is needed.
How it supports your regulatory work
HIPAA. The HIPAA Security Rule requires covered entities and business associates to conduct an accurate and thorough risk analysis and to implement security measures that reduce the risks they identify (45 CFR 164.308(a)(1)(ii)(A) and (B)). The data-flow diagram, findings and test evidence can feed that process. The risk analysis and the decisions in it remain yours.
Changes HHS proposed to the Security Rule in January 2025 are not final, so we do not treat them as requirements (where the proposal stands).
Medical devices. Not every healthcare AI agent is a medical device. Whether software is a device depends on its intended use and indications for use, and some software functions are excluded from the definition (FDA: how to determine if your product is a medical device). For a cyber device, section 524B of the FD&C Act requires the manufacturer's premarket submission to include a plan to address postmarket vulnerabilities, processes that give reasonable assurance of cybersecurity, and a software bill of materials (FDA: cybersecurity in medical devices FAQs). For qualifying device work, we can scope a separate cybersecurity evidence workstream that runs alongside your regulatory and quality process.
What a security review does not establish. It does not demonstrate clinical safety or effectiveness, validate clinical performance, or decide whether a product is regulated. Those determinations belong to your clinical, regulatory and quality teams and their advisers. It is not a HIPAA compliance audit or a certification.
Sources checked October 29, 2026.
Related Medoya Labs work
Medoya Labs projects are our own builds, not client work. Two of them explore problems this review examines:
- Twin Gatekeeper is a proof-of-concept reference architecture in which a policy engine scores each request to a model and decides which checks run before it is answered. It informs how we examine permission enforcement in front of models and agents. Its code is not published.
- Sensitive Data Telemetry Simulator (trazr-gen) is an open-source tool that generates logs, metrics and traces with mock sensitive fields, so a team can check that telemetry is scrubbed before real patient data flows.
Planning an AI deployment that handles health data?
Tell us what the agent connects to, who uses it and the decision or deadline ahead.
