Revising a draft before you approve

Every revision is checked again. You only decide on a version that was checked in the exact state it lands.

You are not stuck with the first thing your AI drafts. Compliance writing is iterative: you want the scope narrower, the wording tighter, a different control behind a requirement. Kanonik is built for that back-and-forth: a revised draft is checked again and comes back to you for your own decision.

Your AI refines before you see anything

When your AI drafts a set of records, it runs each draft through the safety check itself, reads what the check asked for, and improves the draft. It does this a few times at most, on its own, before anything reaches you. Nothing is recorded and no approval link exists while it works.

What reaches you is the finished set, as one approval request. Each item shows How this was prepared: one plain line on what the check asked for and what your AI changed. When your AI could not satisfy the check within its few tries, the item says Needs your judgment instead, and shows what the check flagged on each try. For those items, sending back is usually the better first move.

You can of course talk the work through with your AI before it submits anything. That conversation is where you shape what it drafts. It is not an approval: only your decision on the request is.

Sending an item back

When an item in front of you is not right, choose Send back on that item and say what is wrong, in your own words. Write it the way you would say it to a colleague: "narrow this to production systems" or "the owner is the CTO, not IT". Your note is what your AI works from.

Your AI revises that item. The revision goes through the safety check again and comes back to you for a decision. The old version never lands: once you send an item back, it can no longer be approved.

Send back applies to one item. The other items in the same request stay open for you to decide.

Send back is offered for drafts your AI can revise in place, such as policies, procedures, controls, assets, vendors and risks. For other items, reject the item with a reason and ask your AI to prepare a new version.

Why a revision is always checked again

An approval is tied to the content you were shown. If the content changes, it needs its own check and your own decision. A revised draft cannot ride an earlier approval.

One detail worth knowing: for a link between a control and a requirement, the safety check is judging one thing, whether this control actually satisfies this requirement. If you only reword the note attached to that link, that judgement has not changed, so the earlier result carries over. If you change the control or the requirement, it is a different link and gets a fresh check. For a policy or a control, the check is about the text itself, so any edit to the body is checked again.

Taking time to decide

Between the moment a request reaches you and your last decision, its content is locked. Taking your time does not change what you are deciding on. If the request's hour runs out, Reopen it from To decide: your ticks and decisions are kept, and the content is the same, so the safety check does not need to run again.

Why this matters for your audit

When an auditor asks how a change was vetted, each recorded change carries the result of its checks and the decision of a named person. When you sent an item back, your reason is kept with it. You revised freely, and the record shows who decided what.

More help

Browse every article in the Help center. For anything else, email [email protected] and a person who works on the product answers.