CertifiedData.io

Cryptographic certification for synthetic datasets — prove y

Visit Website
April 7, 2026 How do you actually prevent PII from leaking into test data?

We’ve been bitten by this multiple times—real emails ending up in CSV fixtures and getting committed to repos.

What we’ve learned:

Manual review doesn’t catch it.
“Just be careful” doesn’t work.
Once it’s in a PR, it’s already too late.

The only thing that’s held up is treating it like a build failure:

  • scan in CI

  • fail on high-risk patterns

  • require explicit override if needed

We built a small local CLI to enforce this:

  • deterministic pattern matching

  • no network calls

  • exits non-zero on high-risk findings

Curious what’s actually working for others—especially at scale.

https://github.com/certifieddata/pii-scan

1 Comment

  1. 1

    After dealing with this a few times, the main takeaway for us was:

    This isn’t a review problem—it’s a build-time problem.

    Manual checks miss it, and once it’s in a PR it’s already too late.

    The only thing that’s held up consistently is:

    • CI gating

    • fail on high-risk patterns

    • explicit override if someone really needs to push

    Curious if others have found anything more robust at scale.

About

AI is moving from generating responses to taking real actions, including payments. But when something goes wrong, most systems can’t answer: CertifiedData.io what happened, why it happened, or who authorized it.