Confidentiality and code handling

Handing your codebase to a stranger is the hardest part of buying an audit. Here is exactly what happens to it, including the parts most services leave vague.

What we commit to

Access is limited to the audit

We use a dedicated read-only audit identity. We never ask for your normal account password, and we never connect to your production database.

You choose how your code is processed

The AI-processing mode is your decision, made explicitly at intake. There is no default that quietly sends your code anywhere.

Your source is not used to train our products

We do not use client code to train, improve or benchmark mAIb products.

Your code is never executed on our machines

Static review reads files. If you permit execution, it happens inside a disposable, network-restricted container that is destroyed afterwards.

Temporary materials are deleted

The repository copy, extracted workspace and temporary evidence are deleted at the end of the retention period, and a deletion record is produced.

A mutual NDA can be considered

If your situation calls for one, ask before you purchase.

Your three AI-processing choices

Some audit work is assisted by AI models. Whether that may touch your code is your call, and you make it before we start.

Mode 1 — No external AI processing of my source code

My source code must not be sent to any external AI provider. The audit uses static tools, local deterministic parsers and manual operator review only.

The most restrictive option. Some analysis depth depends on manual review time, so a small number of checks may take longer or be marked as not run.

Mode 2 — Redacted snippets may be processed by a disclosed AI provider

Selected, redacted code snippets may be sent to the disclosed external AI provider to assist analysis. Secrets, personal data and customer data are removed before any snippet is sent.

A middle option. Only fragments relevant to a specific finding leave our environment, after redaction. The provider is named in the report and in our AI-processing disclosure.

Mode 3 — Repository-level AI assistance is permitted

The repository may be processed by the disclosed external AI provider, under that provider's applicable data-processing terms, to assist the analysis. I understand my code is transmitted to that provider under this option.

The fastest and most thorough option. Your code is transmitted to the named provider. We do not make claims about that provider's own training or retention practices beyond their published data-processing terms, which we reference in our disclosure.

What we will not tell you: that “your code never leaves our systems” when you have chosen a mode where it does. Under modes 2 and 3 your code — or redacted fragments of it — is transmitted to the external AI provider we name in our AI-processing disclosure. We also make no claims about that provider's own training or retention practices beyond referencing their published data-processing terms. Whichever mode you choose is recorded, and the report states which one was used.

Retention and deletion

Deleted at the end of retention

The repository copy or uploaded archive
The extracted working copy of your code
Temporary evidence beyond the minimum kept for the report
Any temporary access token, which is also revoked

Kept as a service record

The job record and your consent records
The findings register and the report we issued you
The minimum commercial record for accounting
The deletion record itself

The default retention period is 30 days after delivery. You can request deletion sooner at any point, and we will confirm when it is done. Full detail is in the code-handling policy and the privacy notice.