A HIPAA auditor does not want a description of your security posture. They want artifacts. Hand them a list, and for each item they will ask to see the document, the log, the approval, the proof it operated. HIPAA for AI coding agents is the same exercise, with one new actor in the room: a process that writes and runs code, and whose code can reach systems that hold electronic protected health information.
So start from the artifacts. What do you have to put in front of the auditor when an AI coding agent has access to a database, a cluster, or a service that touches ePHI? That question, not a policy PDF, decides whether the control holds.
The artifacts HIPAA access controls produce
The Security Rule's access control and audit control standards (45 CFR 164.312) expect specific, demonstrable records. Translated into things you physically hand over:
- A record showing each access to ePHI tied to a unique user identity, not a shared account.
- An audit log of activity in systems that contain ePHI, retained and tamper-evident.
- Evidence that access was limited to the minimum necessary for the task.
- Proof that where ePHI was returned to a less-trusted actor, it was de-identified or redacted.
For a human, you assemble these from IdP logs, database audit tables, and ticket trails. For an AI coding agent, the usual sources fail. The agent often runs under one service credential, so unique-user attribution collapses. It logs to its own runtime, so the audit log lives inside the actor you are auditing. Minimum-necessary is meaningless if the agent holds standing access between tasks.
The architectural requirement HIPAA implies
If the artifact has to be trustworthy, it cannot be produced by the agent. An audit log that the audited process can rewrite is not an audit log. So the requirement is structural: the access to ePHI-bearing infrastructure must pass through a boundary the agent does not control, and that boundary must emit the artifacts.
That is what an identity-aware access gateway does. Place hoop.dev between the coding agent and the database or cluster, and the agent's connection is authenticated against your identity provider, scoped just in time, recorded command by command outside the agent, and masked on the way back. One clarification that matters for HIPAA scoping: hoop.dev governs the infrastructure connection, not the model. It does not read the agent's prompt or its generated text. The ePHI it protects is the data flowing over the database and service connections the agent opens.
What you hand the auditor, item by item
- Unique-identity access record. Each agent session authenticates through OIDC or SAML and ties to a named service identity. The gateway's per-session log is your unique-user artifact.
- Tamper-evident audit log. Commands are recorded at the gateway, outside the agent's reach. That is the audit control artifact, and the actor cannot edit it.
- Minimum-necessary proof. Just-in-time, bounded access means the agent does not hold ePHI-system privileges when it is not working. The grant record shows scope and duration.
- Redaction proof. On connections that support masking, the gateway redacts PHI in query results before they reach the agent, and that redaction is part of the session record.
hoop.dev does not hold a HIPAA certification and does not make a connection HIPAA certified; no tool can. What it does is generate the evidence for HIPAA that your covered-entity or business-associate program hands to an auditor.
A worked example
An agent is tasked with fixing a query bug in a reporting service backed by a Postgres database that stores patient records. Through the gateway, it authenticates as agent-reporting, receives scoped read access for the session, runs SELECT statements that the gateway records verbatim, and gets results where the patient name and MRN columns come back masked. When the audit comes, you export that session: identity, timestamps, every statement, masking applied. That is the artifact, complete, without the agent having touched the record.
FAQ
Is an AI coding agent a HIPAA problem on its own?
Only if it can reach systems holding ePHI. The risk is the access, not the code generation. Govern the connection and the agent fits inside your existing HIPAA access controls.
Can hoop.dev deliver HIPAA compliance for us?
No. HIPAA compliance is a program, not a product. hoop.dev generates the access evidence for HIPAA: per-identity logs, scoped grants, and masking proof.
Does the gateway see the agent's prompts?
No. It governs the database and infrastructure connections the agent opens, not the model's input or output.
HIPAA evidence has to come from outside the actor it describes. See the gateway model on the hoop.dev getting started guide, and read how masking and recording work in the open-source code at github.com/hoophq/hoop.
