KryptonGRC Academy
The guide
1What does an AI audit read?
What agents did — their runs, read whole: every turn, every tool call, the final state.
2What is an eval record?
The record of a test run — its dataset, the points it was judged on, called its criteria, and its scores.
3Are judge scores the verdict?
No. Scores are evidence; the auditor’s call is the verdict.
4What does the client do?
Three steps: export, drop, verdict.
5How is a criterion read?
Against the live book — covered, accounted with its reason, or a named gap.
6Can a gap read as covered?
Never. A criterion no framework speaks to is a gap, named.
7Which frameworks speak to testing?
MITRE ATLAS, the NIST AI RMF and ISO/IEC 42001.
8What happens to a failure?
It becomes a named retest in the next record, and it closes on proof.
9How are the courses organised?
By stage of the audit lifecycle — start where it makes sense.
10How many courses per subject?
One — self-paced, with a plain diagram on its card.
11How does a class question run?
One numbered question, one control — you decide first, the reasoning follows.
12When does a module complete?
When its work really happens — evidence first, the stamp second.
13Is there a certificate?
No certificate. Just portfolio.
14Where does the path end?
In your portfolio — your full report and the record of every call you made.
15What is the portfolio proof of?
The work itself — every worked surface carries into the report.
Start where it makes sense. Every course is self-paced, and the path ends in your portfolio.