Making LLM security verdicts verifiable: the evidence gate pattern

Every "AI security analyst" I tried had the same flaw: a correct verdict and a confident-but-wrong...

venerdì 3 luglio 2026 New tab

364 words~2 min read

Every "AI security analyst" I tried had the same flaw: a correct verdict and a confident-but-wrong one are indistinguishable on screen. In security that's not a UX nit — it's the whole problem. So I built USAP around a single rule, and this post is about that rule and three things that fell out of it.

The evidence gate

USAP's output contract is 11 typed JSON fields. The uncompromising one is evidence_references: every verdict must carry at least one source that resolves. Four accepted forms:

mcp:<logical>:<tool>:<call_id> — evidence fetched live from a connected MCP

https://... — a canonical external source (CVE, EPSS feed, MITRE)

Making LLM security verdicts verifiable: the evidence gate pattern

Making LLM security verdicts verifiable: the evidence gate pattern

Related reading

Why evidence matters more than model memory in AI pentesting

Your AI agent's audit trail is not evidence. Here's what makes it one.

The Auditor's AI Workflow: How I Use LLMs Without Trusting Them

Agent Evidence Levels (AEL): grading the evidence your AI agent leaves behind

AI Systems Need Evidence, Not Just Observability

LLM Audits and Guardrails Are Not Enough: Why You Must Filter at the Logit Level

Related reading

Why evidence matters more than model memory in AI pentesting

Your AI agent's audit trail is not evidence. Here's what makes it one.

The Auditor's AI Workflow: How I Use LLMs Without Trusting Them

Agent Evidence Levels (AEL): grading the evidence your AI agent leaves behind

AI Systems Need Evidence, Not Just Observability

LLM Audits and Guardrails Are Not Enough: Why You Must Filter at the Logit Level