Guides
Who May Approve a Firewall Change? Approval Chains and Separation of Duties

A firewall change needs approval from at least one person who is neither the requester nor the implementer, and that approval only counts if nobody involved can change the list of approvers or edit the request after it was approved. In the model below, higher-risk changes need more approvers. An emergency change gets a recorded go-ahead before it is made and a second person's approval within a fixed time afterwards. The record must show who approved which version of the change, and when.
This post is about the approval chain itself. The firewall change management policy template covers the whole policy, and the firewall change management guide covers the process from request to audit trail. Here we take the one part that most often exists on paper and not in the tool.
Is one approver enough for separation of duties?
One approver is enough only if three things are true. The approver is independent, meaning not the requester and not the implementer. The approval cannot be tampered with, meaning the requester cannot edit the approver group, the approval rules or the request after sign-off. And the approver sees the actual rule, not a summary written by the requester. If any of the three fails, the second pair of eyes exists in the workflow diagram only.
NIST SP 800-53 states the principle in control AC-5: "Define system access authorizations to support separation of duties." Its discussion says separation of duties addresses the potential for abuse of authorized privileges, and gives an example that fits firewalls well: the people who administer access control functions should not also administer the audit functions. Control CM-5 adds that organizations should "define, document, approve, and enforce" the access restrictions that apply to changes. Enhancement CM-5(4) calls for dual authorization for implementing changes to organization-defined components.
"Enforce" is the word that decides audits. In our view, a rule such as "nobody approves their own change" has no value if the ticket system lets the requester edit the approver group, because the requester then holds both roles with one extra click. The privilege that matters is not "can click Approve". It is "can decide who may click Approve".
How do approval chains fail in practice?
Approval chains fail at five points. None of them needs bad intent. Each one is a normal convenience that nobody turned off, and each can be tested in a few minutes.
| Weak point | What it looks like | Fix |
|---|---|---|
| Requester can edit the approver group | A project admin adds a colleague, or themselves, to "Approvers" before raising the ticket | Treat any change to an approver group as a High-tier change, approved by someone outside that group and logged |
| Request editable after approval | Destination changes from one host to a /24 subnet; the status still says "Approved" | Bind the approval to a revision of the request. Any edit to source, destination, service, action, window or device list voids it |
| Approver sees the ticket, not the rule | "The app needs the database" is approved; the rule behind it says any | Show the resolved rule in the approval screen, with objects expanded to addresses and ports |
| Change made on the device first | An administrator edits the firewall directly and raises the ticket afterwards | Compare the running configuration with the approved state on a schedule; see drift detection |
| Shared or overlapping roles | One shared admin account; the implementer also holds the approver role | Personal admin accounts, and a tool rule that one person cannot hold approver and implementer on the same change |
How can you test your own approval chain in 15 minutes?
Run the checks below with a test account on a lab firewall or a test ticket queue. Each one should fail or reset. If one succeeds, the chain has a hole at exactly that point.
- Raise a request as a normal user and try to approve it yourself. This must be refused by the tool, not by a reminder in the policy.
- As the requester, try to add yourself to the approver group. This must fail, or open a High-tier change that someone else decides.
- Get the request approved, then edit the destination. The status must drop back to "pending".
- Open the approval screen as an approver. Can you see the expanded rule with real addresses and ports, or only the requester's text?
- Change a rule on the lab firewall with no ticket. The change should appear in the drift report within the interval your policy names.
- Declare an emergency. The tool should demand a named retrospective approver and a deadline before it lets you close the ticket.
Run test 3 first. It is the one that most often turns out to be missing, because ticket tools were built to track work, not to freeze a decision.
Which approval chain fits which risk tier?
Approval should scale with risk: a standard change needs no per-change approver, a normal change needs one or two, and a high-risk change needs three, with at least one from outside the requester's team. The tool should assign the tier from the properties of the rule, and the requester should not be able to pick it. A reviewer may raise a tier but never lower it.
The tiers are our recommendation. The standards ask for the structure, not for these numbers. NIST CM-3 requires organizations to determine and document the types of changes that are configuration-controlled, to review proposed changes and approve or disapprove them, and to document the decisions. The FFIEC IT Handbook separates routine, major and emergency modifications in its change types section. The matrix below turns that into something a firewall team can run.
| Tier | Typical firewall change | Approvers before the change | Extra rule |
|---|---|---|---|
| Standard | Add a host to an existing approved group; any item from the pre-approved catalogue | None per change. Security and the owner approved the catalogue entry once | Someone outside the implementing team samples completed standard changes every month |
| Low | New rule inside one zone, named hosts, named ports | 1: security reviewer | Reviewer is neither requester nor implementer |
| Medium | New rule across a zone boundary; NAT, VPN or routing change | 2: security reviewer and the rule or service owner | Two different people, neither the requester nor the implementer |
| High | Inbound from the internet; any in source, destination or service; access to a regulated data zone; management-plane access; admin accounts | 3: security reviewer, service owner and a change board or security lead | At least one approver outside the requester's team |
| Approval system | Edit to an approver group, an approval rule, a tier definition or the list of firewall administrators | Always treated as High; approver is not a member of the group being edited | Logged separately and reviewed monthly by the policy owner |
| Emergency | Active incident or imminent outage that cannot wait | 1 before: duty manager by phone or chat, recorded. 1 after: a person who did not implement it | Ticket within 24 hours, retrospective approval within 2 working days |
Two details in the matrix are deliberate. The approval system is its own row because it is the tier attackers and shortcuts aim at: whoever controls the approver list controls every other row. And the High tier names an approver outside the requester's team because three people from one team who all report to the same manager are one opinion counted three times.
For the checks that feed the tier decision, such as shadowed, duplicate and over-wide rules, see firewall rule impact analysis.
How should an emergency firewall change be approved?
An emergency change is approved twice: before the change, by a named person who may declare an emergency, and after it, by a different person who checks that the emergency was real and the change was correct. The first approval can be a phone call if it is written down. The second one needs a deadline, because without a deadline it does not happen.
The FFIEC handbook describes the same shape. It tells organizations to specify who can declare an emergency and to designate a group that can meet on extremely short notice to evaluate and approve proposed emergency modifications. It also says management should validate every emergency modification after implementation to determine whether it was correctly classified, and that evaluations and documentation reviews should be completed as soon as possible after implementation. It does not name a number of hours. The 24 hours and 2 working days in our matrix are a starting point you should adjust to your own risk policy.
The same page warns about the abuse case: control should be maintained over personnel who try to get around routine change control by classifying changes as emergencies. That is why the monthly report should show the emergency share of all changes. When the normal path takes two weeks, "emergency" becomes the fast lane. The fix is a faster normal path, not a stricter emergency rule. Our posts on the emergency change workflow and on emergency changes that never get rolled back cover the rest of that path.
What must the approval trail show?
The trail must show who requested, who approved, what exactly they approved, and when, for every change, including emergencies. "Approved" with no name and no timestamp proves nothing in an audit, and neither does a name attached to a ticket that was edited afterwards. The table lists the fields we would insist on.
| Field | Question it answers | Basis |
|---|---|---|
| Requester, with date | Who asked, and when | The FFIEC change request elements include requestor's name and request date |
| Tier and the reason for it | Why this chain applied | Our recommendation |
| Each approver: name, decision, time, comment | Who approved, rejected or asked for more information | FFIEC lists the decision (approved, rejected, more information required, or deferred) and the authorized approver's signature and date; NIST CM-3 asks for documented change decisions |
| The revision that was approved | What exactly they signed off | Our recommendation |
| Implementer, time, configuration before and after | Who did it, and whether it matches the approval | Our recommendation |
| Emergency: who declared it, why, retrospective approver and time | Whether the emergency was real | FFIEC includes a justification for an emergency change on the request form |
| Retention period | How long the record stays available | NIST CM-3 requires retaining change records for an organization-defined period |
The revision field is the one teams leave out. Without it, nobody can show that the rule in the firewall is the rule the approver saw. For the evidence side of this, see firewall change records as legal evidence.
Checklist: is your approval chain real?
- The requester cannot approve their own change, and the tool enforces it.
- The implementer cannot approve the same change.
- The requester cannot edit the approver group; edits to it are High-tier changes approved by someone outside the group.
- The tier is assigned from rule properties, not chosen by the requester.
- High-tier changes have at least one approver from outside the requester's team.
- The approval screen shows the resolved rule, not only the request text.
- Any edit to a field that changes behaviour voids the approval.
- Standard changes come from a catalogue that security and the owners approved, and a person outside the team samples them.
- Emergency changes have a named person who may declare them, a recorded go-ahead, a ticket deadline and a retrospective approver who did not implement the change.
- The emergency share of all changes is reported every month.
- Each record names the approver, the decision, the time and the approved revision, and is kept for the period your regulator sets.
- Direct edits on the device are detected and handled as unauthorised changes.
How we checked this
We read the NIST SP 800-53 Rev. 5 texts for AC-5, CM-3 and CM-5 as reproduced on csf.tools, and the FFIEC IT Handbook sections on emergency modifications and the change request form, on 21 November 2026. We did not quote PCI DSS in this post. Its requirement text sits in the PDF on the PCI SSC document library, and the policy template links the PCI requirements it maps. The approval tiers, the 24-hour and 2-working-day limits, and the 15-minute test are our own recommendation, built on many migrations in which old rule bases had to be justified line by line.
FwChange builds this chain into the firewall change workflow: tiers from rule properties, approvals bound to a revision, and one audit trail. To see which of your current evidence an auditor would accept, run the free NIS2 Readiness Check.


