Trust and security

Security at Bunker.

This page is written for the person doing the security review, not the person doing the buying. It states what is in place today, and it is equally clear about what is not.

Last updated 1 September 2026.

How access works

Bunker does not create a second identity system alongside yours. It runs inside the one you already control.

Governance and audit

The governance layer is the product, not a setting. It runs server side, where a user cannot turn it off.

Writes are governed, and fail closed

Any action that changes one of your systems goes through a policy check on the server. If the policy engine cannot reach a verdict, the action does not run.

Approvals ledger

Actions that need a human decision are held and recorded, with the requester, the target and the outcome.

Audit ledger

Governed activity is recorded against the user and the organization that performed it, and administrators can export it.

Per user metering

Model usage and cost are metered per user and per organization, so an administrator can see exactly who consumed what.

Where your data lives, and what we hold

Bunker runs on Microsoft Azure in the East US 2 region.

Subprocessors

The complete list of third parties that process customer data. Microsoft services inside your own tenant are not subprocessors, because that data never leaves your control.

SubprocessorPurposeLocation
Microsoft AzureApplication hosting, database, secret storage and operational loggingUnited States, East US 2
AnthropicModel inference. Prompts and the content sent with them are processed to generate a responseUnited States
CloudflareDNS and network edge for tuatarabunker.aiGlobal edge network

The model provider

Your prompts and the content sent with them are processed by Anthropic to generate a response. That is worth being precise about, because it is the question every security reviewer asks next.

Those certifications are Anthropic's, not ours. Bunker does not inherit them, and we are not going to imply otherwise. They tell you the model provider underneath us is independently audited. They say nothing about Bunker's own controls, which are described above and are not yet independently attested.

What this does not cover. Anthropic processes the content of a request in order to answer it. If a prompt contains data your policy says may not leave your environment, that policy is the control, not this page. Governed write actions and connection scoping are enforced by Bunker. What a user chooses to type is not.

Compliance status

Stated plainly, because you will find out anyway.

Bunker does not hold a SOC 2 report today. We are working toward one, scoped to the Security trust services criteria. We are not going to give you a date we cannot stand behind, and we would rather tell you that now than at contract stage.

What we will do instead, today

Answer a security questionnaire, including the CAIQ. Walk your security team through the architecture. Put our data handling commitments in the contract rather than in a brochure.

Proof of concept data

For a time boxed evaluation we will commit in writing to deleting your data at the end, naming what is deleted, and confirming when it is done.

Reporting a vulnerability

If you believe you have found a security issue in Bunker, tell us and we will respond.

[email protected]

Please include enough detail to reproduce the issue. We will confirm receipt and keep you updated until it is closed.