@epilot/anonymization is a data-protection component: weaknesses in it can directly affect the privacy of end customers. We treat every report seriously.

• Preferred: use GitHub's private vulnerability reporting on this repository ("Security" → "Report a vulnerability").
• Please do not open public issues or pull requests for security problems, and do not include real personal data in reports — synthetic examples only.

We will acknowledge reports promptly, keep you informed, and credit reporters who wish to be credited once a fix is released.

Beyond the usual (dependency issues — there are no runtime dependencies — or code execution), we explicitly consider the following in scope:

• Masking bypasses: any input on which a value classified as PII survives anonymization (e.g. a value shape that escapes the entity walker, an address/payment structure that leaks a field, a pattern the scrubber should catch but doesn't).
• Determinism/derivation weaknesses: anything that makes pseudonyms reversible or cross-org linkable without the secret (e.g. insufficient input domain separation).
• Fail-open behavior: any code path where an error or missing configuration results in raw values passing through instead of redaction.

• The inherent properties documented in the README's Limitations section (pseudonymization is re-derivable with the secret; non-strict heuristics are best-effort; quasi-identifier re-identification in small populations).
• Security of the secret storage itself — the secret is injected by the consumer and never handled by this package beyond HMAC input.

The latest published minor release receives security fixes. The curated classification defaults are versioned data: security-relevant additions to them are released promptly as minor versions.