Adrian DunkleyAI Risk Management

Dependencies and global outages

Find the shared provider behind your separate backups

Use the Cloudflare outage to test whether your fallback works when a shared service fails.

On 18 November 2025, Cloudflare reported widespread problems delivering core network traffic. Cloudflare’s Matthew Prince wrote that the “feature file, in turn, doubled in size” after a database permissions change. Start with the operator’s account when you write the exercise. Keep early explanations in the chronology as unconfirmed or ruled out. Use verified evidence in the final cause field. Matthew Prince, Cloudflare outage report

Trace the full route to a customer

List the services a customer or employee touches, then follow each dependency. A main portal and a status page might depend on the same identity provider or traffic service. A backup call centre might still use the same application to view a policy. Ask an owner to verify each link from architecture notes, vendor records or an observed test. HM Treasury, The Orange Book

Give AI the approved inventories and ask it to find shared names, missing ownership and suspected reuse. Treat matches as questions. Common branding, similar provider names or a single reference in a ticket can mislead the model.

Test the fallback while the primary service is unavailable

Practice fallbackDependency to checkEvidence to collect
Public status pageSeparate path to publish and view updatesRecord of a test message viewed externally
Call-centre scriptAccess to approved offline guidanceA staff member retrieves the right version
Manual incident logReconciliation when the primary tool returnsEvery field and event ID imports once

In an exercise, record when staff notice the problem, when the duty manager switches channels and when the first customer can complete the service. Ask which people remain blocked. Keep a manual decision log that can later be reconciled with the primary record.

Copy this prompt

Using the approved service inventory, availability requirements and fallback plan, produce a dependency map by business service. List provider, owner, supporting evidence and any unverified link. Identify where a primary and fallback route may fail together. Draft a discussion-based provider outage test with a decision owner and success criteria for restoring access. Do not assume services are independent where evidence is missing.

Put the result on the risk register

Record the suspected shared dependency, tested workaround and remaining exposure. Ask the service owner whether to fund a separate route or accept a reduced service mode. Use NARM if an automated service or AI agent can widen the failure.

Sources and references