AI with guardrails

AI that obeys your roles, your hand-offs, and your law

A Compliance Pre-Check before every AI document action. Tracked changes for every author, including the AI. Access that mirrors the user. The guardrails aren't settings; they're how the AI works.

Compliance Pre-Check

Before every AI action, not after

Every AI request that would modify a document runs through a jurisdictional pre-check first. The document's jurisdiction (country + state if applicable) and document type drive the analysis. If anything you asked for would breach a statute, the AI shows you what's at risk before any change lands.

What an issue looks like

Worked example: "Reduce the notice period to 7 days."

Field Example
Area Notice Period
Statute Employment Relations Act 2000, s.67
Risk level High — illegal / void Medium — likely unenforceable
Issue A 7-day notice period falls below the statutory minimum for this employment type and jurisdiction.
Suggested alternative Use 14 days minimum for fixed-term roles, or align with the role-specific minimum in this jurisdiction.

You always get three options — never blocked, always informed:

Proceed anyway

Apply the change as requested. Decision recorded in the audit trail.

Apply only the safe parts

Compliant portions of your request go through; risky portions are skipped.

Cancel

Discard the change entirely.

Runs before every AI action with zero setup NZ, AU, and US statutory minimums supported Included in every plan, even Free Trial
Tracked changes universal

Every change is a redline. Yours, theirs, the AI's.

Every edit is recorded as a tracked change with author and timestamp. AI edits are tracked even with track-changes toggled off in the toolbar; there is no way to make AI changes look like human ones. The AI never accepts its own changes; you always decide which redlines stand.

  • Author and timestamp on every redline, human or AI
  • AI changes always tracked; toolbar toggle does not affect them
  • AI never auto-accepts; you accept or reject every line yourself
  • Per-line accept/reject is independent of the document-level Approve/Reject decision

Pending AI redlines block more AI work in the same area

If the AI has already proposed changes in a section that you haven't accepted or rejected yet, the AI won't pile new edits on top in that same area. You stay in control of what the AI built before it builds more.

Why this matters: it prevents AI-on-AI cascades where each new request silently builds on un-reviewed prior changes.

AI inherits user access

The AI has exactly the user's access — never more

No grant → the AI can't reference the file. View-only grant → the AI can read but never edit. Edit grant → the AI can act on it. Same chat, two users, two different answers, based on each user's own access.

Compliance Guard isn't a setup. It's how the AI works.

Zero configuration. Included in every plan. The pre-check, the tracked changes, the access boundary, all on by default, all of the time.

AI that's safe to ship in a law firm

Pre-checked. Permission-aware. Tracked. Auditable. By default, on every plan.