The accounts team expects an automatic exchange to finish every evening. After a security change, the website still opens but the exchange stops. The obvious response is to ask someone to let the accounting programme through. Before doing that, it helps to understand where a request goes and which computer is meant to accept it. Otherwise a quick repair may restore the exchange by also reopening a route the team intended to close.
A server is a computer or service that answers requests. A browser—the programme you use to open websites—can ask it for a page. An accounting programme can ask a service to accept invoice data and return a result. A protective intermediary, often called a proxy, can receive these requests first, check them and forward accepted requests to the server behind it. The answer returns through the intermediary. 17 11
The server behind the intermediary is called the origin: it is where the application or its content lives behind the intermediary. Protecting it means deciding which connections it should accept, not merely hiding its address. Cloudflare’s guidance describes several ways to do this. The choice depends on how the service is built and which legitimate connections it needs. 11
Imagine the intended route is accounting programme → protective service → origin → answer. If the origin also accepts the same request directly from the internet, that second route can miss checks made only by the protective service. Changing the normal route has not by itself instructed the origin to refuse other connections. The origin needs its own rules about what may reach it, while the application still needs to decide what each accepted request may do. 11 12
Knowing the address does not mean the door opens
A report may say it discovered an origin address. That is not the same as showing that the server accepted an unwanted request. Network restrictions or other checks may still reject it. Cloudflare recommends reviewing old and current ways addresses may be disclosed, but address secrecy alone is not the access rule. 11
An internet address tells other computers where to attempt a connection. It does not promise that the receiving computer will accept it. A known address can therefore coexist with effective access restrictions. Conversely, an address that is not advertised today may still be known from earlier use. Cloudflare’s guidance discusses historical records as well as current disclosures. Concealing an address can reduce exposure, but accepting or refusing a connection is a separate control. 11
Nor is every separate route a mistake. An approved partner connection may use a different route with its own protection. In that case, the right question is how that connection is recognized and controlled. A separate entrance can be intentional; an unexplained, unprotected entrance is a different matter.
Consider an accounting program that sends data automatically each evening. After a protection change, the homepage still opens, but this exchange stops. A test of the homepage would miss the business failure. Cloudflare’s bot documentation warns that legitimate automated or mobile application traffic can be challenged. Normal work therefore belongs in the change test, not just attempts to block unwanted visits. 9 10
Three ways to control the connection
1. Limit who can connect
One approach is to accept network connections only from specified sources, such as the protective service and approved partners. This narrows who can reach the origin. The permitted sources need to match the intended arrangement; otherwise the rule may block a legitimate partner or allow more than intended. Cloudflare describes this as an origin-protection option, not a substitute for the application’s own checks. 11
These restrictions are commonly enforced by a network firewall: rules that permit or refuse connections according to the configured conditions, such as where the connection comes from. This is different from a web application firewall that examines web requests. Refusing an unwanted connection early can keep it from reaching the web programme at all. But permitting a protective network only identifies that allowed source; it does not automatically identify the website operator’s particular Cloudflare account or the invoice owner. 11 12
There are also two connection segments to consider: from the visitor or accounting programme to Cloudflare, and from Cloudflare to the origin. Protection on the first segment does not describe the second segment’s settings. 12
Encryption makes information on a connection unreadable to outsiders in transit. Checking a credential—a digital proof accepted by the receiving service—has a different job: deciding whether the connecting service is an accepted source. The origin needs to enforce the check; simply asking Cloudflare to send a credential does not make an origin that ignores it reject everyone else. 12
2. Check a certificate at the receiving server
Another approach asks the connecting service to present a digital credential, called a client certificate. The origin checks it before accepting the connection. Cloudflare calls its feature Authenticated Origin Pulls, or AOP. It requires the relevant configuration on the origin as well as enablement at Cloudflare. 12
The connection between Cloudflare and the origin must also be encrypted: the supported settings are named Full and Full (strict), not Off or Flexible. The setting’s name alone does not show that rejection is actually enforced. 12
There is an important detail about what that certificate identifies. Cloudflare’s global AOP certificate is shared across accounts. It establishes that a request came from Cloudflare’s network, not specifically from your account. Where the design needs that narrower distinction, the documentation describes using your own certificate for the relevant website or named service. Neither arrangement identifies which customer is entitled to a particular invoice; the application still has that job. 12
3. Start the connection from inside
A third approach changes how the connection is made. With Cloudflare Tunnel, software on the origin establishes outbound connections to Cloudflare, without requiring the origin to have a public address to which internet traffic can be routed. This is a different arrangement from accepting ordinary incoming connections with the AOP certificate check. The documentation says AOP does not apply to origins connected through Tunnel, whose connections have their own authentication. These are not boxes to tick indiscriminately on the same checklist. 11 12
“Outbound” describes who starts the connection. Instead of leaving the origin waiting for new internet connections, the installed Tunnel software starts a connection from inside towards Cloudflare. Requests can then reach the service through that established arrangement. It is not a claim that the application no longer receives requests or needs customer permissions. It changes how the route to it is established and authenticated. If another unwanted route remains open elsewhere, selecting Tunnel does not automatically close that unrelated route. 11 12
Let the right software in for the right reason
Return to the evening accounting exchange. If it is stopped, the team needs to know whether a network restriction, a certificate check, a bot control or the application itself caused the failure. “Security blocked it” is too vague to choose a safe correction. Different controls make different decisions.
Ask the team to trace the intended exchange using its own records: did the protective service receive it, did the origin receive it, and did the application return the expected result? Each answer concerns a different step. A browser challenge before the origin and an application refusal after it need different corrections. Likewise, an approved partner connection may have a separate route whose own controls should be preserved. The goal is to correct the failed step without removing protections unrelated to that failure.
A program’s self-declared name is not a suitable replacement for proving which program is connecting. Requests can include a label called User-Agent, often describing the browser or software. Such labels can claim another identity; MDN documents this problem. Letting everything bearing a friendly software name avoid protection would confuse a label with authentication. 15
A supported exception should skip only the intended check for appropriately identified traffic, leaving unrelated controls in place. Cloudflare’s Skip feature can omit selected checks, but does not work identically across products: it supports Super Bot Fight Mode, not Bot Fight Mode. The operator must identify the actual feature before choosing the exception. 13
For a temporary repair, I recommend agreeing when the exception ends. Otherwise a workaround for one failed connection can become an undocumented permanent opening. Also decide in advance how to restore the previous approved settings if the change disrupts business; restoring service should not silently mean opening the origin to everyone.
Test both the closed door and the working service
The acceptance test has two parts. The unwanted route must be refused, and the intended work must succeed. Ask for an authorized exercise with test accounts and invented records that covers the relevant browser visits and automated connections. Do not experiment with another company’s live server or real customer invoices.
A short result should say what was tested and what was not. If the public website, customer portal and invoice-delivery service are separate, a result for one cannot stand in for all of them. The business does not need private network details or secret credentials in an email; it needs a clear explanation of the routes being protected.
Before approving the change, ask: “What should stop working, what must keep working, and how will we undo the change safely?” Those questions make protection and continuity part of the same decision, rather than leaving the accounts team to discover the difference the next morning.
Sources
[9] Cloudflare: Bot Fight Mode
[10] Cloudflare: Super Bot Fight Mode
[11] Cloudflare: Protect your origin server
[12] Cloudflare: Authenticated Origin Pulls
[13] Cloudflare: Custom rule Skip action