“Your provider has an exposed API.” In a security report, that sentence can sound as though somebody has left your invoices on the street. But it may mean only that the report found an address used by customer programs. To interpret the finding, we need to separate reaching a service from receiving its confidential records.
You already do something similar when you visit a bank’s login page. The page must be reachable before you sign in. Reaching it does not, by itself, give you access to somebody’s account. This comparison explains only that distinction; it does not tell us whether an invoice provider’s controls are actually correct.
What is reachable, and what is being requested?
An accounting program can send a message directly to a provider’s program asking for work or information. The outgoing message is a request; the returning answer is a response. The agreed way the programs organize these exchanges is called an API. An address used for one of its operations may be called an endpoint. You can understand it as the place to which the program sends a particular kind of request. 13
Suppose the service supports requesting an invoice copy. The customer program sends a request identifying the invoice it wants and presents the access information required by the service. The receiving software must establish the requester’s permissions and decide whether that requester may obtain this document. Only then should it return the permitted information. Knowing the address for requesting copies is different from having permission for a copy. Permission needs to be checked for the requested action on the requested record. 1
A public instruction page might describe this operation so that accounting-software developers can connect their products. Reading those instructions tells us that the operation is documented. It does not demonstrate that the operation works without access checks. Equally, documentation promising a check does not demonstrate that every deployed version correctly carries it out. Versions and their protection require continuing attention. 7
Consider three imaginary report entries. “Documentation address found” concerns a page of instructions. “A request without the required access information was refused” concerns that request’s response. “An authorized test account received a test document belonging to a different fictional company, contrary to the agreed rule” concerns a tested permission failure. The third is much more direct evidence of a permissions problem. Calling all three simply “exposed API” throws away the distinction that helps a business decide what to do.
A report’s method explains what it can know
Imagine the report author first reads an address listed in another organization’s public index. The author is looking at an earlier observation collected by somebody else. The author has not thereby sent a fresh request to the invoice provider. The address may still be useful, but its history matters.
Next imagine the author opens the provider’s public documentation page. A browser is the program used to open web pages. Opening the page makes the browser ask the provider for it, and the provider responds. That is direct contact with the provider, even if the page needs no login. It remains a request for documentation, not a test of invoice permissions.
Watching ordinary work in your own authorized account answers another question: what occurred during that account’s activity? Deliberately altering messages to ask whether the service will release a document outside the account’s rights is a separate, active test. A reachable address is not permission to do that, and a small number of attempts does not supply authorization.
For a simple public review, I would agree in advance which documentation or status pages may be read and keep direct requests few and read-only. I would exclude login attempts, searches for hidden operations, changes to document references, requests that create or change records, and attempts to create load. Those are proposed limits for an agreed review, not instructions to investigate any service you can reach. If real sensitive information appears unexpectedly, stop rather than explore it further.
Read the actual answer, not only the colour
Some reports concern information attached to an answer rather than an invoice inside it. This accompanying information is called response headers. Certain headers instruct a browser about handling a response. Whether a particular instruction is relevant depends on what was returned and how it is used. API security guidance explicitly discusses applicable headers for browser-facing APIs; it does not make every missing browser instruction proof that invoices are accessible. 8
Here is a simple interpretation example. A checker receives an ordinary error page and inspects it. It may accurately report that a named header was absent from that page. If the checker could not receive or inspect the answer, however, it cannot turn that gap into “the header is missing”. The difference is between observing absence and having no observation.
An error can itself disclose too much information, such as sensitive internal details that should not be sent to the requester. That is a legitimate configuration concern. But it should be described in relation to the actual answer: which details appeared, in what response, and why they matter. An error page and a response containing a confidential invoice do not demonstrate the same problem. 8
The reverse shortcut is equally misleading. One refused request does not establish protection for every operation, account and version. It shows that this request was refused. Old interface versions can remain relevant after a newer one is introduced, so the supplier needs an accurate account of which versions are deployed and protected. 7
Choose the follow-up that matches the question
Suppose the report finds public instructions for an old interface version while a customer notice says that version has been retired. If your accounting program uses it, the useful follow-up is to clarify the applicable instructions and any necessary change. That inconsistency is not evidence of a break-in. It is a documentation and version-management question. 7
If the real concern is whether customers can retrieve each other’s invoices, ask for a different kind of demonstration. The provider can use expressly authorized, separate test facilities with invented companies, invented documents and temporary accounts. Agree the allowed and refused outcomes first. Keep the exercise small, and check resulting records or files as well as the answer on screen. Do not guess references in the live service or use real customer invoices.
Finally, keep track of which system was examined. A warning from the marketing website may prompt a question about care, but that website and the customer API may run on different software. Neither a weak website result nor a clean one resolves invoice permissions. Ask the report author for two ordinary sentences: “We examined this and observed this”; then “This question was not examined.” You can then choose a repair or a further authorized review for the actual uncertainty, rather than treating every visible address as a leak.
Sources
13 — MDN: Overview of HTTP — client/server requests, responses and sessions
1 — OWASP API1:2023 Broken Object Level Authorization
7 — OWASP API9:2023 Improper Inventory Management
8 — OWASP API8:2023 Security Misconfiguration