As of 13 August 2026, AI can only partly check your business firewall rules.

What to actually do

1. Hand it to a person

   The route this page recommends

   A person who owns the outcome does this end to end, worth it when the failure is dear.

2. Use a tool built for this

   Second choice

3. Do it yourself

   The distant third

   A chat interface, power-user skill, and roughly 1 hour until you can act on the result.

   How to actually do it

   1. Open the firewall management console and export the current rules, including rule order, source, destination, service, action, schedule and logging settings, without including credentials or secret keys.
   2. Open the current network diagram, asset list and service documentation, then gather the business purpose of each important connection, including supplier access, remote access, public services and internal administration.
   3. Redact passwords, API keys, private personal data and unnecessary public IP details from the export, while preserving rule identifiers, ranges, ports, protocols, order and comments.
   4. Paste the prompt and the redacted export, network notes and relevant denied or allowed traffic examples into an AI chat, then ask it to keep the configuration read-only and separate evidence from assumptions.
   5. Copy the findings into a review table and compare every claimed conflict or exposure with the original rule order and the firewall manufacturer's current documentation.
   6. Ask a network or security colleague to confirm the intended traffic flows, investigate every low-confidence finding and reject any recommendation whose operational impact is unclear.
   7. Test only approved, reversible changes in a non-production path or controlled maintenance window while monitoring allowed traffic, denied traffic and service availability, then record the final decision and approver.

   Prompt

   Act as a cautious firewall-review assistant, not an administrator. Review the firewall configuration and network context below for security weaknesses, accidental exposure, redundant or shadowed rules, overly broad source or destination ranges, unsafe ports and protocols, ordering problems, missing logging, and rules that appear stale. Do not invent the purpose of a rule, claim that a port is safe without evidence, or recommend deleting or changing a rule without stating the risk and the information needed first.
   
   Return:
   1. A short summary of the highest-priority findings.
   2. A table with rule identifier, observed behaviour, concern, evidence from the supplied text, confidence, and safer next question or action.
   3. Possible rule conflicts, including which rule is evaluated first and why.
   4. Items that cannot be assessed without live traffic, packet captures, asset ownership, vendor documentation or a network diagram.
   5. A reversible validation plan that does not change production, including what to monitor and what would make the test unsafe.
   6. A list of questions for the person responsible for the network.
   
   Treat the configuration as read-only. Separate observations from assumptions. Do not provide credentials, secrets or personal data in your response. Do not tell me to apply a change until a qualified network or security professional has reviewed it.
   
   Firewall platform and version: [platform and version]
   Environment and business purpose: [brief description]
   Network diagram or traffic-flow notes: [paste or describe]
   Firewall rule export: [paste the export]
   Relevant logs or denied-traffic examples: [paste redacted examples]
   Known services, suppliers and required traffic: [list]
   Maintenance and testing constraints: [describe]

   Open it prefilled in ChatGPT or Claude, or copy it into Gemini, which takes no prefill link.