How AI is changing WordPress security

A notice about a vulnerable plug-in calls for a concrete decision. Are you using that plug-in? Is your version affected? And is a security update available? Anyone who manages a WordPress website gets such questions regularly. AI is meanwhile changing the research behind those notices: security professionals can use it to examine software, but attackers are also gaining more powerful tools.

Een geografische toegangsbeperking op de server kan een deel van het ongewenste verkeer weren. Zo’n geofence of landenblokkade bepaalt welke verbindingen worden toegelaten. Daarna blijft de vraag wat iemand via een toegelaten verbinding kan doen, bijvoorbeeld als een plug-in een beveiligingslek bevat.

A geographic access restriction on the server can block part of the unwanted traffic. Such a geofence or country block determines which connections are allowed. After that, the question remains what someone can do through an allowed connection, for example if a plug-in contains a security flaw.

Even with geoblocking, security updates and secure accounts remain essential

The country rule limits where a connection may come from. It does not fix a software bug and does not tell you whether an allowed visitor is trustworthy. AI is changing the possibilities for investigating and abusing flaws, which means following up on notices continues to require attention.

Why you may see more vulnerability reports

More published vulnerabilities do not automatically mean that all the involved flaws only recently emerged. Existing code can be examined again, with different tools or more researchers. Discovering, publishing, and actually exploiting are different events.

On 10 April 2026, Wordfence offered a look into its own bug bounty program. According to the company, the share of reports in which researchers themselves indicated they were using AI had risen from 16% in the first full measurement week to about 66%. Registration began at the end of November 2025. This concerns self-reported AI use at Wordfence, not a measurement of all vulnerabilities or attacks within WordPress.

For you as an owner, what matters most is whether a report affects your installation. A general news item does not yet say that your website is vulnerable or hacked. To determine that, you need to look at the relevant software, versions, and conditions.

What Wordfence Argus shows

On 27 August 2026, Wordfence described its internal research system Wordfence Argus. The company uses this system to investigate serious vulnerabilities. Among other things, Argus combines AI models with orchestration, verification, and other software. It is not an additional WordPress plug-in that a website owner needs to install.

A concrete example followed in the publication of 1 September 2026 about a vulnerability in Gravity Forms. Wordfence reported that Argus had discovered the issue on 9 August. Under certain conditions, the flaw could allow unauthorized file uploads; the impact depended in part on the form settings and server configuration. According to the publication, the flaw was fixed in version 3.0.3, released on 20 August.

In this case, the remediation update was available before the publication of the research article. That gave website administrators the opportunity to fix the flaw. For such a report, check the vendor’s current security information and use a supported version with the necessary fixes. The version numbers mentioned describe this specific example from 2026.

Why the same progress can help attackers

The possibilities of AI are not limited to security researchers. In its threat assessment through 2027, published in May 2025, the UK’s National Cyber Security Centre expects that AI will, among other things, support vulnerability research and the exploitation of known flaws. The centre expects this could leave less time between the disclosure of a vulnerability and its exploitation.

That is a well-founded threat expectation, not proof that every attack on a small business website is already being carried out by an autonomous AI agent. Even without AI, automated scans and attacks exist. A traffic spike or a foreign IP address does not tell you which tools are behind it.

The practical consequence is that you need to organize the follow-up of security notifications. An available update only helps once someone determines that it is needed and also implements it. Therefore, define who assesses notifications, who can intervene, and how you verify after a change that the website continues to work. This way, security becomes part of regular website maintenance and the follow-up does not remain dependent on news items read by chance.

What geofencing on a server can protect

With geographic access restrictions on a server, it concerns the estimated origin of an internet connection. For this application, the terms geoblocking and GeoIP blocking are also used. A firewall can refuse requests from certain countries or allow only requests from selected regions.

This can be useful when permitted access is clearly delineated. Think of an administrative environment used by a small team. The right trade-off depends on the website and the services that need to connect to it.

A country rule assesses the geographic classification of an IP address. It does not tell you the identity or intent of the visitor. An allowed connection therefore is not yet a reliable user. Moreover, this is about server access; geofencing in, for example, vehicle tracking uses physical zones for a different application.

Why geoblocking does not fix software vulnerabilities

Suppose your server receives a request from an allowed region. That request can then access a vulnerable part of the website. Whether the country rule worked correctly says nothing about how the plug-in processes that input. Both security layers assess something different.

Attackers can also route traffic through an allowed country. Digitaal Vlaanderen mentions a VPN as a way to bypass geoblocking and advises using the technique as an additional security layer.

The same applies to a stolen administrator account. A geographic restriction can block certain connections, but it does not independently verify whether the right person is using a password. That is why secure accounts, software updates, and appropriate access rights remain necessary alongside a geofence.

What each security layer does

A web application firewall, abbreviated WAF, examines requests for characteristics of attacks. That is a different function than a country filter. Wordfence describes how its WAF filters malicious requests. The protection depends, among other things, on the available rules and the configuration. These measures complement each other.

Measure	Helps with	Limit of the protection
Geoblocking	Restrict access based on geographic origin	Does not fix a software vulnerability
WAF	Detect and block certain attack patterns	Coverage depends on rules and configuration
Security update	Fix a specific software vulnerability	Does not automatically remove previous misuse
Two-factor authentication	Extra verification when logging in	Does not fix a flaw that can be exploited without logging in
Backup	Restore files and database	Does not prevent a break-in and must be restorable

What to check when you receive a new security notice

Assess a security notice together with your host or webmaster. These questions help you move from a general message to an appropriate action for your website.

1. Are you using the affected software and version? Compare the exact plug-in or theme with your own installation. Similar names may refer to other products.
2. Under what conditions is the vulnerability relevant? Read whether, for example, an account, a certain feature, or a server setting is required. Do not draw conclusions from only the headline of a message.
3. Is there a recovery update or workaround? Follow the vendor’s advice. If there is no solution, have it assessed whether a feature should be temporarily disabled or the software replaced.
4. Can you change and recover in a controlled way? Ensure you have a usable backup and, after an update, test key functions such as forms and payments.
5. Is there reason for further investigation? An update fixes a flaw, but does not prove that earlier misuse has been undone. If you see suspicious signals, also review logs, accounts, and modified files.

The official WordPress hardening guide covers the broader fundamentals, including updates, secure access, logging, and recovery. For a practical overview for business owners, you can also consult our guide on securing your website.

Do you see unknown administrators, strange redirects, or changes that no one on your team has made? Treat that as a potential incident and involve your host or webmaster. Then follow the separate step-by-step plan for recovering a hacked website. Just updating or restoring a random backup is insufficient as long as the cause and scope are unclear.

Keep customers and essential services reachable

Even for a website that mainly serves Belgian customers, legitimate traffic can come from other countries. Think of a customer traveling, or an external payment or monitoring service. When making changes to country rules, test which functions and connections are truly necessary.

Search engines also deserve attention. According to the technical requirements of Google Search, Googlebot must be able to reach the content and receive a successful response in order to index it. An error message that only one other crawler receives does not in itself prove that Google is being blocked.

For public online stores, the European rules against unjustified geoblocking also apply. Therefore, assess a restriction of the entire store differently from a restriction of admin access. Let the concrete situation be a factor.

Allowing crawlers requires a separate decision

An AI crawler that fetches public content is different from AI that helps an attacker investigate software flaws. Decide separately which crawlers may use your content and how you limit abuse of your website. Cloudflare distinguishes, within AI traffic, among other things search, agents, and training. Blocking such a category is not a substitute for fixing a security vulnerability.

Frequently asked questions about AI and WordPress security

Do you still need to update if you already use geoblocking?

Yes. A country rule limits connections based on origin. A security update fixes a flaw in the affected software. Even via an allowed connection, someone can try to exploit an existing vulnerability.

Do more vulnerability reports mean WordPress is becoming less secure?

Not necessarily. More intensive research can also make existing flaws visible. Therefore, look at which software was examined, which versions are affected, and how quickly fixes become available. The number of reports alone does not provide a complete risk picture.

Are all attacks with AI automatic and autonomous?

No. AI can support individual steps, while people continue to make choices and assess results. How much autonomy a system has differs per application. A suspicious request does not show whether AI was used.

Does a Belgian website have to block all foreign visitors?

That does not follow from the fact that you mainly serve Belgian customers. Customers can travel and external services can connect from other countries. Assess the necessary access per application and test the consequences of each restriction.