How MxGuard's hybrid AI filtering works

MxGuard uses a two-layer approach so that speed and accuracy work together.

1. Machine-learning classifier (the fast layer). A gradient-boosted ML model scores the easy 95% of mail in milliseconds — sub-5ms decisions. Obvious good and obvious bad mail is handled instantly.
2. Claude LLM evaluation (the careful layer). The borderline messages — the ones that matter most — are handed to Claude, Anthropic's large language model, which reads them the way a person would before a verdict is reached.

Transparent verdicts. Every decision comes with a plain-English reason, so you can always see exactly why a message was tagged, quarantined or rejected. There is no black box: the console shows you the reasoning behind each verdict.

This combination is what lets MxGuard catch the carefully written, well-targeted attacks that slip past older filters, while keeping legitimate mail flowing without delay.