Security at MedicalAgentOS

Security is designed into every layer of MedicalAgentOS — from tenant isolation in the data model to guardrails around every AI agent action.

Platform security

• Encryption in transit (TLS 1.2+) and at rest (AES-256) for all customer data.
• Multi-tenant architecture with organization- and practice-level isolation enforced server-side.
• Role-based access control with fine-grained permissions; no privilege inferred from role names alone.
• MFA, secure session management, and configurable session timeout policies.
• Secrets management with KMS-backed key rotation; no credentials in application code.
• Continuous monitoring, security event alerting, and audit-log integrity protection.
• Encrypted backups with tested recovery procedures.

AI-specific safeguards

• All patient-provided content is treated as untrusted; defenses against prompt injection and tool abuse.
• Agents never infer permissions from natural language — authorization is server-side only.
• Per-skill permission checks, autonomy levels, and human-approval requirements.
• A dedicated guardrails engine for risk classification, emergency phrase detection, and escalation.
• Every model call carries purpose, trace ID, and validation of structured outputs.
• Cross-patient and cross-tenant context leakage prevention tests in CI.

Responsible disclosure

We welcome reports from security researchers. Email security@medicalagentos.com with details; we acknowledge within 2 business days and do not pursue action against good-faith research.