Security at Atticus

Our approach

Protected health information is the most sensitive data there is. It can't be reset like a password or reissued like a credit card — once it leaks, it's leaked for life. We built Atticus knowing that from day one, so security is built into the architecture, not bolted on afterward.

Atticus operates as a HIPAA Business Associate to health plans, ACOs, and providers. Every safeguard described on this page applies to every member we serve, on every engagement, with no premium tier required.

Data protection

• Encryption everywhere. All data is encrypted at rest with AES-256 using managed keys (AWS KMS), and in transit with TLS 1.2 or higher.
• PHI isolation. Health information lives only in an encrypted clinical data store. It never sits in message queues, caches, logs, or analytics systems.
• Opaque identifiers. The systems that move work around — queues, schedulers, event buses — carry only random, meaningless IDs. Even if one of those layers were exposed, there would be no names, no dates of birth, and no clinical data in it.
• Minimum necessary. Services query only the fields they need, never whole records by default.

Access & auditability

Access to systems and data follows least privilege: people and services get the narrowest permissions that let them do their job, and nothing more. There are no shared accounts and no standing broad access.

Every read, write, and deletion of health information is logged to an append-only audit trail. The trail is tamper-evident — each entry is cryptographically chained to the one before it — and its integrity is verified automatically every day. If a record were ever altered after the fact, the chain would break and we would know.

Infrastructure

Atticus runs on Amazon Web Services under a Business Associate Agreement with AWS. Development, staging, and production environments are fully segregated, and real member data exists only in production. We use continuous monitoring and managed threat detection across the environment, with alerting on anomalous access patterns and configuration drift. Secrets are stored in a managed vault, never in code.

AI safety guardrails

Atticus uses AI agents to run care management. They operate inside hard limits:

• Agents never prescribe. They do not recommend specific medications or dosages — ever. Clinical decisions belong to clinicians.
• Identity confidence thresholds. When an event can't be matched to a member with high confidence, it goes to human review instead of triggering an action. A wrong-patient alert is a patient-safety event, so we would rather stay silent than guess.
• No PHI in model prompts. Models work from de-identified summaries and structured references, not raw member records.
• Humans can take over. Any AI conversation can be taken over by a person at any time.
• Everything is logged. Every AI action lands in the same tamper-evident audit trail as everything else.

Vulnerability management

We run regular security audits across the codebase and infrastructure, scan dependencies for known vulnerabilities on an ongoing basis, and patch promptly. Pre-commit and CI controls block secrets and unsafe patterns before code ever ships.

Responsible disclosure

If you believe you've found a security vulnerability in an Atticus product or service, we want to hear from you. Email security@atticushealth.com with enough detail for us to reproduce the issue. We will acknowledge your report, keep you informed as we investigate, and never pursue good-faith research.