What makes a website secure

Secure means there is as little as possible to attack, what remains is patched inside the window attackers now use, and there is a tested way back if it fails.

Quick scan

  • Patchstack measured a median of five hours from disclosure to first exploit attempt for the WordPress flaws attackers cared about, yet the UK's Cyber Essentials scheme allows 14 days for critical and high-severity fixes.
  • A static site still has things to attack, namely the hosting account, the domain, the GitHub account, the build tools, their code libraries and any third-party script.
  • Ask whoever looks after your site how many days a serious fix takes, when a backup was last restored, and what grade Mozilla's HTTP Observatory gives it.
Read the full article5 minutes
In this article

A website is secure when there is as little as possible to attack, when what remains is patched faster than attackers move, and when there is a tested way back if it fails anyway. An owner can check each part. Here is what they mean, how fast attacks now come, and what to ask whoever looks after your site.

What can be attacked

Every website has a front door, the pages visitors see. Most business sites also have a back door, an admin login, an application that builds pages on demand, and a database behind it. The back door is where nearly all successful attacks go, because that is where the code runs and the passwords are.

For a site on WordPress, the back door is large. In 2025, 11,334 new security flaws were disclosed across WordPress, its plugins and its themes. Nine in ten were in plugins. Nearly half, 46 per cent, had no fix available on the day they were made public. Those are Patchstack’s figures. Patchstack sells WordPress security, and Wordfence, another vendor, counts in the same range.

A static site has no application running on the public server and no database to query. The pages are built in advance and served as files. The editor your team uses is reached at an address on the site, but the page there is itself a static file that runs no code on the server; signing in goes through GitHub with two-factor authentication, and a stolen password there cannot execute anything on your site. What is left to attack is the hosting account, the domain, the GitHub account, the tools that build the site, the code libraries those tools depend on, and any script loaded from a third party. That is a much shorter list, but it is not an empty one, and we say so.

How fast attacks come

The window between a flaw being published and someone trying it on your site is now hours, not weeks.

VulnCheck found that 29 per cent of the vulnerabilities first exploited in 2025 were exploited on or before the day their public record appeared. Mandiant’s 2026 report puts the average time to exploit at minus seven days: on average, attacks begin a week before the fix is public. For WordPress specifically, Patchstack measured a median of five hours from disclosure to the first exploit attempt for the flaws attackers cared about. Cloudflare has logged an attempt 22 minutes after a proof-of-concept was published.

The UK’s Cyber Essentials standard, which many larger UK buyers now require of suppliers, asks for critical and high-severity fixes within 14 days. US law sets no deadline for a private business. Even 14 days is slower than the attackers, which is why the number of things needing patches matters more than how diligently they are patched.

Most of this is automated. Attackers scan every address on the internet for a known flaw and exploit whatever answers. In the UK government’s 2026 survey, 43 per cent of businesses identified an attack or breach in the previous year, and 46 per cent of small businesses.

The part that is not on the server

In 2025 the attacks that made the news were not on websites. They were on the tools that build them. In March 2025 a build tool used by more than 23,000 code repositories was altered to steal secrets. In September 2025 a self-spreading worm compromised hundreds of published code packages that developers install without reading. OWASP, the body that lists the most serious web application risks, moved supply-chain failures to third on its 2025 list.

So a site with no server-side application still has a supply chain, the code libraries it is built from, the build service, and any script it loads from someone else’s server. We pin dependencies to known versions and update them deliberately, keep third-party scripts to the minimum, and run the build with the least access it needs. That does not make the risk zero. It makes it small, known and watched.

The way back

Every site, however built, needs a tested way back. That means backups that exist somewhere other than the server they back up, that are taken often enough that losing the gap would not hurt, and that have been restored at least once to prove they work. A backup nobody has restored is a hope.

For a static site, the whole site and every change ever made to it lives in a version-controlled repository in your name or transferable to you on request, and rolling back is a one-line change. For a site that stays on WordPress, it means database and file backups on a schedule, kept off the host, with a restore drill in the calendar.

What secure means on this site

When we say a site we build is secure, we mean all of the following, and each can be checked.

  1. There is no application or database on the public server. The pages are files. The editor’s sign-in page is a static file too, and the sign-in itself happens at GitHub with two-factor authentication.
  2. Every connection is encrypted, with the modern protocol versions only, and the browser is told never to connect without encryption. A free scanner such as Mozilla’s HTTP Observatory grades this; we build to grade A.
  3. The response headers tell the browser what the page may load and where it may be shown, so a stray script or a forged frame is refused.
  4. The site’s dependencies are pinned and updated on purpose, within 14 days for anything rated high or critical, and within 48 hours where a public exploit exists.
  5. Editors sign in through an account with two-factor authentication and can change content, not code.
  6. The whole site is in version control, in the client’s name or transferable on request, and a restore has been tested.
  7. Forms go to a service that holds the message and nothing else, and the site loads no third-party script it does not need.

Four questions to ask whoever looks after your site

  1. Does our public website run an application that checks passwords on our own server, and who has an account? (On WordPress, add /wp-login.php to your address and see what appears.)
  2. When a serious flaw is announced in something we use, how many days until it is fixed on our site, and how would you know it had been announced?
  3. When did we last restore a backup, and how long did it take?
  4. Scan our site at developer.mozilla.org/en-US/observatory and tell me the grade.

A good answer to all four is possible on any platform. A vague answer to any of them is the thing to worry about.

Where this definition is used

The Websites page builds to it. WordPress migration moves a site from the large back door to none. Ongoing care is the patching, the backups and the restore drill as routine work. The Website Risk Analysis checks a site you already have against every line of it and gives you the immediate actions in writing.

If you cannot answer the four questions about your own site, that is the place to start.

  • NextWhat makes a website safe for the people who use itSafe is about the visitor. It means the site collects nothing it does not need, can be used by anyone, and does what it says, and each of those is checkable wherever your buyers are.
  • PreviousWhat makes a website fastFast is a measurement, not a feeling. Your page is usable on a mid-range mobile phone in under two seconds for most real visitors, and you can check it yourself in five minutes.

Sources (10)

  1. State of WordPress Security in 2026 Patchstack (vendor), 25 February 2026, covering calendar 2025. Accessed 28 October 2026.
  2. State of Exploitation 2026 VulnCheck (vendor), 21 January 2026, covering 2025. Accessed 28 October 2026.
  3. M-Trends 2026 Google Cloud, Mandiant, 23 March 2026, covering 2025 investigations. Accessed 28 October 2026.
  4. Application Security Report 2024 update Cloudflare (vendor), 11 July 2024. Accessed 28 October 2026.
  5. Cyber Security Breaches Survey 2025/2026 Department for Digital, Culture, Media & Sport, and Department for Science, Innovation & Technology, 30 April 2026, fieldwork August to December 2025, 2,112 businesses. Accessed 5 November 2026.
  6. Cyber Essentials, Requirements for IT infrastructure v3.3 National Cyber Security Centre, April 2026. Accessed 5 November 2026.
  7. OWASP Top 10:2025 OWASP, 2025. Accessed 4 November 2026.
  8. tj-actions/changed-files supply chain attack (CVE-2025-30066) Wiz (vendor), March 2025. Accessed 28 October 2026.
  9. Shai-Hulud npm supply chain attack Wiz (vendor), September 2025. Accessed 28 October 2026.
  10. HTTP Observatory tests and scoring MDN, Current. Accessed 28 October 2026.