Clone Solutions
Template

Clone Governance Charter

The rules every Clone works under: purpose, permitted data and actions, what it must never do, approval gates, escalation and how autonomy is earned.

What this is for

An agentic AI rollout needs rules that a security team can sign, a business owner can understand, and an auditor can test. This charter is that rulebook. Part A sets the organization-wide rules: who decides, how autonomy is earned, what no Clone may ever do, and how incidents are handled. Part B is a one-page record for each Clone: its purpose, the data it may use, the actions it may take at its current autonomy level, and who it escalates to. Clones add capacity to teams. The charter makes sure people stay accountable for decisions.

When to use it

Draft Part A in the Week 3 governance design session, after the Connections & Permissions Plan and Security Configuration Baseline are agreed. Add a Part B record for every Clone in the Design week of each department rollout. Update a Part B record at every autonomy promotion. Review Part A quarterly at the business review.

Who's involved

  • InfoSec lead: co-owns the charter; approves every Part B record and every promotion.
  • Executive sponsor: approves Part A and resolves escalations that reach steering.
  • Business owners: own the Clones in their departments and co-sign promotions.
  • Legal or compliance: reviews Part A where regulated work is in scope.
  • GRC Specialist and GRC Clone: draft and maintain the charter.

How to fill it in

  1. Agree the governance roles in Section A1. Use names, not teams.
  2. Adopt the autonomy ladder (A2) and set promotion criteria (A3). Make criteria numeric where you can.
  3. Write the prohibited actions (A4). Start with the defaults and add your own; do not remove defaults without InfoSec approval.
  4. Set risk tiers and the minimum human oversight for each (A5).
  5. Define escalation and incident handling (A6, A7) and link to your existing incident process.
  6. For each Clone, complete a Part B record before it is connected to any production system.
  7. Get sign-off. Publish Part A to everyone who uses Clones; publish Part B records to the teams that use each Clone.

The template

Part A: Organization-wide rules

A1. Governance roles

RoleResponsibilityName
Executive sponsorFinal authority on scope and risk appetite
InfoSec leadApproves connections, Part B records and promotions; can pause any Clone
Business owner (per department)Accountable for outcomes and for every action their Clones take
Approver (per workflow)Approves actions at Execute with approval
Org adminApplies configuration; cannot approve their own changes
Champion (per department)First line for user questions and feedback
GRC Specialist / GRC CloneMaintains this charter, the risk register and evidence

A2. The autonomy ladder

LevelNameWhat the Clone doesWhat a person doesMinimum oversight
1AssistDrafts, summarizes, researchesEverything else: every send, every decisionPerson reviews every output before use
2RecommendPrioritizes and proposes next actions, with reasoningDecides and actsPerson decides on every item
3Execute with approvalActs after a named person approves each actionApproves, edits or rejectsNamed approval on every action
4Execute within limitsActs on its own inside set thresholds; escalates anything outsideOwns thresholds, reviews exceptions, audits a sampleWeekly exception review; sample audit
5OrchestrateCoordinates multiple workflows and Clones across systemsSets goals, monitors, can override at any momentContinuous monitoring; quarterly review with InfoSec

Autonomy is set per workflow, not per Clone. One Clone can draft customer replies at Assist and route tickets at Execute within limits.

A3. Promotion criteria

A workflow moves up one level at a time. Promotion requires the evidence below and sign-off from both the InfoSec lead and the business owner, recorded in an Autonomy Promotion Evidence Pack.

PromotionMinimum time at current levelVolume of evidenceAccuracyOverride / rejection rateIncidents
Assist → Recommend2 weeks100 items or agreed≥ ___% acceptable without major editsn/aNone above severity ___
Recommend → Execute with approval4 weeks200 items or agreed≥ ___% recommendations accepted≤ ___%None above severity ___
Execute with approval → Execute within limits6 weeks300 approvals or agreed≥ ___% approved unchanged≤ ___% rejectedNone above severity ___; clean audit log review
Execute within limits → Orchestrate1 quarterAgreed per workflowSustained at targetExceptions within agreed bandReviewed at the quarterly business review
Example: AP matching, Recommend → Execute with approval4 weeks900 invoices96% of proposed matches accepted4% overriddenNone above severity 3

Demotion: any incident above the agreed severity, or accuracy below target for two consecutive weeks, moves the workflow down one level until a review is complete. Either the InfoSec lead or the business owner can demote alone.

A4. What no Clone may ever do

Defaults (keep all unless InfoSec approves a change):

  • Approve its own actions, or approve another Clone's actions where a human approval is required
  • Create, change or delete user accounts, permissions or its own connections
  • Move money, change bank details or release payments without a human approval at that moment
  • Sign, accept or terminate a contract on the organization's behalf
  • Make a final decision on hiring, firing, compensation, credit or any decision with legal effect on a person
  • Give legal, medical or financial advice to customers as if from a licensed professional
  • Share Restricted data outside the systems and people approved for it
  • Contact customers, regulators or the press on a channel that has not been approved
  • Disable logging, change its own instructions or bypass an approval gate
  • Present itself as a human when asked

Organization-specific additions:



A5. Risk tiers

TierDescriptionExamplesHighest autonomy allowed without steering approvalReview cadence
LowInternal, reversible, no sensitive dataMeeting briefs, policy Q&A, internal researchExecute within limitsQuarterly
MediumInternal impact or Confidential data; reversible with effortTicket triage, invoice matching, CRM updatesExecute within limits, with thresholdsMonthly
HighCustomer-facing, financial, regulated or Restricted dataCustomer replies, payment runs, patient or employee dataExecute with approvalWeekly at InfoSec sync
ProhibitedAny action in A4Not permittedn/a

A6. Escalation

SituationClone actionEscalates toResponse time
Outside thresholds or low confidenceStops and asksApprover for the workflowSame business day
Missing or conflicting knowledgeSays so; does not guessChampion2 business days
Possible policy breach or sensitive data exposureStops; flagsInfoSec lead1 hour
User asks for a prohibited actionDeclines and explainsBusiness owner (logged)Next weekly review

A7. Incidents and the kill switch

  • Named people who can pause any Clone: InfoSec lead, org admin, business owner (own Clones).
  • AI incidents follow the existing incident response process. Severity scale: link or define 1–4.
  • Every incident is logged in the AI Risk Register with root cause, fix and any demotion.

A8. Change control

  • Production Clones run pinned skill versions.
  • Any change to instructions, knowledge sources, skills or connections is tested against the Test Scenario Pack before release.
  • Changes to High-tier workflows need InfoSec approval before release.

Part B: Clone record (one per Clone)

FieldEntry
Clone name
Role (hired into)
Department and business owner
Purpose (one sentence)
Users and channelse.g. AP team in Teams; email intake
Knowledge sources
Permitted data (by classification)
Connections used (IDs from the Connections & Permissions Plan)
Risk tierLow / Medium / High
Escalation owner
Prohibited actions beyond Part A
Version / skill pins
Date approved

Workflows and autonomy

WorkflowCurrent levelApprover (if level 3)Thresholds (if level 4)Target levelNext promotion review
Match invoice to PO and receiptRecommendn/an/aExecute within limitsWeek 10
Post matched invoice in ERPAssistn/an/aExecute within limits for variance ≤ 2% and invoice below agreed amountAfter matching reaches level 3

Good looks like

  • Autonomy is per workflow. Charters that grant a whole Clone "level 4" hide the risky actions inside the safe ones.
  • Criteria are numbers. "Reliable" is not a promotion criterion; "96% accepted over 900 items" is.
  • Prohibited actions are specific and tested in the Test Scenario Pack.
  • Two signatures on promotion, one signature to demote. Making it easy to step back is what makes it safe to step forward.
  • Part B exists before the connection. No production access without a signed Clone record.
  • Common mistake: writing a charter nobody reads. Publish Part A in plain language to every Clone user.

How Clone helps

The GRC Clone drafts Part A from your existing policies and the pre-read, creates a Part B record for each Clone from its Design Spec, and tracks promotion evidence against the criteria in A3. The GRC Specialist facilitates the governance design session and reviews every change. In the portal, the charter is the Clone Governance Charter deliverable under Security & GRC; each Clone's current autonomy level, next review date and sign-off history are visible to InfoSec and business owners.

Sign-off

RoleNameDateDecision
InfoSec leadApproved / Approved with conditions
Executive sponsorApproved
Legal / compliance (if in scope)Reviewed
Business owner (Part B, per Clone)Approved