Threat-informed vs taxonomy-informed: why your guardrails need a model

Two ACSM platforms will both claim “custom guardrails for your stack.” Read the small print. There's a real difference between guardrails generated from a taxonomy (a catalogue of categories) and guardrails generated from a threat model (a map of the risks your repo actually has). One scales to your stack; the other scales to your stack and your trust boundaries.

Taxonomy-informed: the floor

A taxonomy-informed platform takes a published catalogue (OWASP Top 10, CWE, the Cisco AI Security Taxonomy) and matches it against your stack. The output: every Django app gets the Django Django guardrails. Every FastAPI app gets the FastAPI guardrails. Useful, fast to set up, and the floor every ACSM platform has to clear.

Threat-informed: the ceiling

A threat-informed platform reads your repo for entry points, trust boundaries, sensitive data flows, and the AI failure modes most likely to show up in your stack. The same FastAPI app gets different guardrails depending on whether the JWT middleware was hand-written or imported, whether the SQL driver is sync or async, whether your endpoints serve internal traffic only or face the public internet.

Why the difference shows up in PR noise

A taxonomy match generates a lot of rules — most of them don't apply to the line that just landed. A threat-informed match generates fewer rules — but every rule that fires fires on a line that actually crossed a real threat. The PR review surface goes from “the scanner found 40 things” to “VibeReview flagged this one line because it's exposed to untrusted input.”

How to ask vendors about it

The interview question: “Show me two of your customers' guardrail sets, side by side. If you handed me both, could I tell which customer each one belonged to from the rules alone?” If the answer is no, you're looking at taxonomy-informed. If the answer is yes, you're looking at threat-informed.