WordPress Security Survey 2026: What 67% Incidents Mean

The WordPress security survey 2026 found a 67% incident rate. Here's what that means for 2FA, AI threats, and breach recovery planning.

A small business owner and an IT consultant sitting side by side at a cluttered office desk, comparing printed login records next to an open laptop and a notebook

The Number at the Centre of the Melapress WordPress Security Survey 2026

Melapress has been asking WordPress professionals the same uncomfortable questions for four or five years running, and the WordPress security survey 2026 arrived with a figure that will get quoted in a lot of posts over the next twelve months: 67% of respondents said they had dealt with at least one security incident in the past year. Two out of every three people who filled in the form. That is not a small sample of unlucky hosts. It is a majority of the people who care enough about security to answer a security survey in the first place.

Robert Abela founded Melapress and has spent 25 years in IT security, starting in antivirus software and web security scanners before moving into WordPress. His company does not sell firewalls or malware scanning. The plugins cover activity logs, two-factor authentication, passkeys, and user visibility, which is the layer after someone has already logged in. That background shapes which questions get asked. He talked through the results with Nathan Wrigley on the WP Tavern Jukebox podcast (opens in new tab), and the conversation is worth listening to in full if you want the raw responses rather than my summary of them.

The respondent mix matters more than the headline. Agencies, enterprise security teams, freelancers, and product owners all answer the same survey, and they do not mean the same thing when they tick the box marked “incident”. A freelancer whose client’s brochure site got defaced for six hours and an enterprise lead who spent a week on a credential-stuffing investigation are both counted once. A near miss that someone noticed at the last minute may sit in the same bucket as a full database exfiltration. The survey does not publish severity weightings, and I would not pretend the 67% figure tells you your own risk level. It tells you what a self-selecting group of WordPress professionals reported, which is a different and still useful thing.

So this post does not treat 67% as a verdict on your site. It treats it as a prompt to go and look at your own numbers, because those are the only ones you can actually act on.

Background: how the WordPress security survey 2026 gets its data

The survey runs on a questionnaire Melapress distributes through its own list and the wider WordPress community, and it has been going long enough now that the trend line matters as much as any single year. Robert spent years as a systems engineer and QA on antivirus products before moving into web security scanners and product management, then founded Melapress a decade ago. That path shows in the question design. Someone who has spent years thinking about what happens inside an authenticated session tends to ask about user accounts, login events, and audit trails rather than firewall rules or WAF tuning.

Running for four or five consecutive years gives you a comparison point a one-off poll would never have. If 67% reported an incident this year and a similar share reported one two years ago, the better question becomes why that number has not shifted despite more tooling being available to everyone. I do not have last year’s exact figure in front of me, and I am not going to invent one. That is the gap you accept when you are working from a podcast conversation rather than the underlying spreadsheet.

What the questionnaire cannot do is verify anything. Every incident is self-reported, and nobody is cross-checking whether the defacement a respondent remembers came from an unpatched plugin or from a cache rule someone botched on the host. Memory is unreliable across twelve months, and the people who answer are the people who already care about the subject. If your agency had a quiet year, you might skip the survey entirely or fill it in without much thought. If you spent March cleaning up a compromised membership site, you will answer every question with precision and volunteer details nobody asked for.

Just as important is who never appears in the sample. Shared hosts, most plugin authors, and the enormous population of site owners who have never subscribed to a security newsletter are largely absent. Nobody surveys the person who runs a five-page site for a dental practice and has not logged into wp-admin since the last content edit. The 67% figure comes from a population more engaged than average, and that should sit next to the number every time you quote it.

Half of what makes this survey useful is watching how the same questions get answered differently year over year by roughly the same kind of respondent. The other half is knowing what it structurally cannot tell you.

WordPress security survey 2026: direct answers to the questions people ask

How many WordPress sites actually had an incident in 2026?

Respondents put it at 67%, which is the headline Robert Abela brings to the WP Tavern episode (opens in new tab). The word doing all the work there is “incident.” The survey does not weight severity or type, so a defaced homepage on a site with four visitors and a customer table walked out the back door both land in the same column. When Abela talks through the findings, he is describing a bucket rather than a threat level, and the bucket is the point. A 67% figure tells you how many people had a bad day, not how many had the same bad day.

Does anyone have a breach recovery plan?

Here the gap opens up. Concern about security runs high across the sample, and a written list of what to do at 2am, with names, credentials, and restore points on it, mostly does not exist. That matches what I see on my own boxes. I could tell you which plugin I would disable first, but I have never actually written it down, and neither has most of the sample. Think about what a one-page plan replaces. Right now it is you, half awake, scrolling through your host’s support chat while a customer emails to ask why the checkout page says the database is unavailable.

Is AI making WordPress attacks harder to defend against?

The survey points to more vulnerabilities surfacing and surfacing faster. Abela’s read is that AI has changed the volume and the pace rather than the underlying mechanics, and he is careful not to claim every site is equally exposed. A plugin with a public CVE and a working proof of concept gets probed within days. A site behind a maintenance contract with a patch schedule waits out most of that noise. Where the effect lands hardest is on the sites nobody is watching weekly.

Do 2FA and activity logs change the outcome?

Partly, and the survey does not pretend otherwise. Authentication hardening shows up as something respondents have adopted and trust, which makes sense given how much of the questionnaire circles that ground. Activity logs answer the question after the fact, which is who logged in and what they touched, not whether that login should have been allowed. Neither one fixes a vulnerable plugin, and neither one keeps a copy of your database from last Tuesday.

What is happening now

The same incident means different things depending on what the site does. An e-commerce store with live carts and card data fears downtime and data loss first, because both of those convert directly into refunds, chargebacks, and a merchant account review. A membership site with recurring billing fears account takeover, since a stolen session is a paying customer gone. A brochure site with five pages and a contact form mostly fears defacement and the phone call from the owner who noticed the homepage was advertising something else. The survey reflects that spread rather than a single consensus ranking, which is the honest answer. Anyone who tells you WordPress site owners agree on their top risk has not asked enough of them.

The finding I keep coming back to is the distance between awareness and mitigation. Nearly everyone in the sample knows what a breach looks like and can name the usual entry points. Far fewer have changed a password policy, restricted admin access, or written anything down. That gap is more useful than the 67% figure because it is the part you can actually close on a Tuesday afternoon with no budget.

There is a pattern I recognise from my own tickets. Experienced WordPress users, the ones who have been running sites for a decade, still patch reactively. Not because they do not know better, but because a plugin update is a change, and a change on a working site is a risk they would rather defer until the site is already broken. So the security patch lands in the same maintenance window as the emergency restore. Less technical users show the opposite bias. They worry most about the dramatic scenarios, a hacker posting on their blog, their customers’ details appearing somewhere, and they reach for a plugin they saw recommended rather than the boring fix, which is admin account hygiene and 2FA on the two accounts that matter.

Both reactions leave the same hole open. The survey does not measure severity, as I said earlier, so treat the risk rankings as sentiment rather than a threat model. What it does measure convincingly is that most respondents are reacting to security rather than scheduling it.

What it means in practice

A breach recovery plan for a two-person team should fit on one page and be readable at 2am by somebody half awake. Mine has four lines: who calls the host, which backup I restore from and where those credentials live, how customers get told, and who takes over if I am the one offline. That is the whole document. Anything longer turns into an incident response document nobody opens, and I have watched agencies with genuinely good intentions write fifteen-page runbooks that were never once consulted during a real outage.

Vulnerability management at this scale starts with admitting you cannot monitor everything. Pick three sources you will genuinely read: your host’s security notices, the changelogs of the plugins you actually run, and one vulnerability feed. Turn the rest off. A monitoring inbox with 400 unread alerts is worse than no monitoring, because it trains you to ignore the message that mattered. If you have WP-CLI available, wp plugin list --update=available on a Monday morning gives you a real list in about two seconds, and that beats a dashboard you never open.

2FA and activity logs sit at different points in the timeline. Two-factor authentication stops credential stuffing and password reuse at the door, which is most of what hits small sites. It does nothing about a malicious plugin, a compromised hosting account, or a stolen session cookie. Activity logs pick up on the other side of login: a new administrator, a role changed from subscriber to editor, a plugin installed at 3am. Activity logging cannot prevent any of that, and if an attacker reaches admin they can often clear the log or disable the plugin, which is why shipping logs somewhere offsite is worth the extra step. Neither tool substitutes for the other.

Training keeps appearing in the responses as the thing people mean to do. Intent without a calendar entry is just a good intention. Book thirty minutes, cover password managers and phishing, and repeat it in six months. That is a better return than another plugin.

What to expect next

My expectation, and I want to be clear this is a read on the data rather than something the survey proves, is that the discovery side keeps accelerating. An AI-assisted pass over a plugin’s code will surface an input sanitisation gap faster than a human reading the same file, and the summary Robert Abela gave on the WP Tavern podcast (opens in new tab) points at a volume of newly surfaced issues rather than a handful of extra reports. Once a bug is written up, working proof-of-concept code tends to follow within days instead of weeks. That compresses the window between disclosure and exploitation, and it lands hardest on plugins with one maintainer who patches after their day job.

Whether vendors can keep pace is a separate problem from whether they are trying. A host with a security team can absorb a busy month. A plugin shop run by two people cannot, and the queue in front of them grows whether or not they answer it. I would expect more small plugins to get abandoned or quietly forked, which pushes the maintenance burden onto whoever runs them. For a site with fifteen plugins, that is fifteen small bets on fifteen strangers.

On which findings I would trust again next year: the gap between awareness and mitigation. That one looks structural, not seasonal, and I do not see what would close it without someone assigning the work to a named person. The 67% figure I would treat more carefully. It will move, but a lot of that movement will come from who chooses to answer. Agencies and hosts are overrepresented in a survey distributed through the WordPress professional community, and a freelancer running forty client sites reports incidents at a different rate than someone with one brochure site. Unless the respondent mix stays roughly stable, year-on-year comparisons will tell you more about the mailing list than about WordPress.

I would also not expect the AI numbers to line up year over year. The questions around AI threats are still being written, and the tooling keeps shifting under the people answering them. Next year’s report is worth reading for the direction, not for a delta. Keep your own incident log running so you have a baseline that does not depend on anyone’s sample.

Start here this week

Nothing in the previous section changes what you do on Monday morning. The useful work is small and unglamorous, and it fits inside an hour if you stop reading and start clicking.

Pick one site you actually own. Not the client site you keep meaning to hand off, and not the one your host manages for you. If you have WP-CLI installed, this prints every admin account on the install:

wp user list --role=administrator --fields=ID,user_login,user_email,user_registered

The registration dates do most of the arguing for you. I have run that command on inherited sites and found accounts belonging to contractors who left two years ago, plus an address on a domain that no longer resolves. Delete what you cannot justify. Downgrade the rest to editor and watch whether anyone complains. If nobody notices within a week, the admin role was never doing any work.

Then turn on two-factor authentication on that same site, for every account rather than just yours. Activity logging goes on at the same time, because 2FA tells you whether someone got in and the log tells you what they touched afterwards. Neither step takes long. Doing both before you read another security article matters, since the article will convince you the problem is larger than it is and you will close the tab having changed nothing.

The recovery plan is one page, and it does not need a template. Who you call when the site is down, which backup you restore from and where that backup lives, and the two sentences you send customers if the database is gone. Save it somewhere you can reach when the hosting dashboard is the thing that is broken, because that happens, and your notes should not live inside the outage.

Set the calendar reminder now, for the same week next year. The point is not to reread the Melapress report but to put your own numbers next to it. Four dated lines in a notes file about what broke and what you did will tell you more about your sites than any percentage in someone else’s sample.

So put the reminder in your calendar before you close this tab, then go spend the hour on the site.