If you believe you have found a security issue in Axiom — including an evasion technique that defeats its detection — please report it privately rather than opening a public issue. Open a GitHub security advisory on this repository, or contact the maintainer directly. Please include what you observed, how to reproduce it, and what you expected instead.

This repository is a technical specification, not a deployable implementation. Reports about the specification are welcome; there is no running service here to test against. Particularly valuable:

• a described mechanism that would not work as documented
• an architectural weakness the threat model does not already acknowledge
• anything in this repository that should not have been published — a value, identifier, threshold or pattern that slipped past sanitization That last category is the most useful report we can receive.

This repository withholds detection calibration on purpose: thresholds, limits, confidence cut-offs, backstop patterns, classifier prompts and scheduled timings. Documentation of a detector is a manual for evading it. The architecture is published so the design can be evaluated; the calibration is withheld so it cannot be circumvented. Requests to publish specific values will be declined.

Reporters are credited unless they prefer otherwise.