You type a company’s web address, press Enter and wait for its page. What reaches your computer is not usually a photograph of the finished screen. Your browser is the program you use to open websites. A program is a set of instructions a computer follows. The browser receives material and assembles what you see. 25

The website’s computer supplies that material over the internet. A computer providing this service is called a server. Your browser and the server therefore have different jobs: one asks for and presents the page, the other answers the request. 25

The exchange has a structure. The browser sends a request: in effect, “Please send the page at this address.” The server sends a response. Along with the page’s content, the response carries small named fields describing it or giving handling instructions. These fields are called HTTP response headers. HTTP is the set of communication rules used for this request and response, not a separate app you need to install. Headers are normally hidden from the page view because they are meant for software rather than the reader. 26

Some headers concern ordinary delivery, such as identifying what kind of content arrived. Others have a protective purpose and are called security headers. The browser reads the instructions it understands and applies them to relevant behaviour. Different instructions control different things; their names on a report are not interchangeable badges of quality. 9 10

Why an encrypted connection is only the beginning

A page can contain several kinds of material. Its main document describes the text, headings, links and places for images. This description is written in a format called HTML. Other instructions specify colours and layout; these are called style sheets. A program can give the page behaviour, such as opening an invoice preview after a click. Such a browser program is called a script, commonly written in the language JavaScript. You need not read any of these formats to understand that displaying text and running a program are different actions. 25

For example, the invoice page may tell the browser where to get its logo and the program for its viewer. The browser can then make further requests for those pieces. What looked like one page arriving is often several responses working together. That matters for security: a handling instruction can govern a particular response or the page’s use of other material, depending on the instruction. The fact that the homepage sent one does not mean every response from the business does. 26

HTTPS adds encryption to the connection: it makes the travelling information unreadable to someone merely watching the connection and protects it against undetected alteration in transit. But encryption does not decide whether received instructions should be executed. A page delivered through a protected connection can still contain a programming mistake. Browser handling rules address what happens with the content, rather than replacing protection of its journey. 5 6

The same invoice screen may therefore need several protections at once. None is redundant merely because another protection exists.

Return to the program opening our invoice preview. The portal can send a policy limiting which scripts may run. That policy is called Content Security Policy, or CSP. Another instruction, HSTS, tells the browser to keep using HTTPS for a specified period. One limits activity within a page; the other concerns the connection to the site. 5 6

These settings do not decide whether you are entitled to read a particular invoice. The application still has to make that decision. The distinction is useful: a browser can follow its instructions correctly while the application makes a separate mistake about access.

“Treat this as the kind of file I said it was”

A website labels the content it sends. It can say, for example, that a response is plain text, an image or a script. That label is called the content type. The browser sometimes has ways to infer a type from the material itself. A security instruction known as nosniff limits that guessing. For scripts and files defining a page’s appearance, it can make the browser reject content delivered with an unsuitable type. 9

Here is the practical idea. A page asks for a program, but the response is labelled as something else. Rather than trying to make it work anyway, the browser should respect the mismatch. The website must therefore supply both the correct label and the protective instruction. Adding the instruction while leaving the labels wrong can stop legitimate parts of the page working. 9

You do not need to remember the full header name, X-Content-Type-Options. What matters is the relationship: the site identifies what it is sending, and the browser is told not to reinterpret it in ways the instruction forbids. This does not inspect an invoice for fraud or decide who may download it. 9

“How much of this address should the next site learn?”

Following a link can involve more than requesting the destination. The browser may also provide information about the page you came from. This is called referrer information. A setting called Referrer-Policy controls how much of that referring address accompanies relevant requests. It can limit the information or omit it. 10

Suppose a customer page’s address contains a customer reference. A link from that page to another service creates a sensible privacy question: should that service receive the full referring address, only the site’s origin, or no referrer at all? Here, the site’s origin is its base address, identifying the connection type and website rather than the individual page. Technically it also includes the connection’s port number, which distinguishes services at an address and is often not written out in ordinary web addresses. The policy controls this particular route of disclosure. It cannot prevent every other way a page might send information. 10

The simpler design lesson is not to put secrets in page addresses unnecessarily. A setting that reduces one form of disclosure is useful, but it is not a reason to make the underlying address more revealing.

What a missing-header warning really tells you

A report saying “missing security header” needs a little translation. Which instruction is missing, and from which page? A check of the public homepage does not automatically describe the pages employees see after signing in. Nor does it describe a separate system that exchanges invoices between businesses.

Presence also needs interpretation. CSP can be used in a report-only mode: the browser can report activity that would break a proposed policy without blocking it under that policy. This helps developers introduce changes, but it is not the same as enforcing the proposed restriction. 6

The useful next question is short: “What behaviour would this instruction change on the page we actually use?” An answer such as “It stops wrongly labelled scripts from running” teaches you something. “It improves our score” does not. Once the purpose is clear, the developer can make the change and check that the page still works. You can judge the explanation without becoming the person who configures the website.

Sources

[5] IETF: RFC 6797 HTTP Strict Transport Security

[6] MDN: Content-Security-Policy

[9] MDN: X-Content-Type-Options

[10] W3C: Referrer Policy

[25] MDN: How the web works

[26] MDN: An overview of HTTP