Skip to content
EN
Log in Start now
Netslim
Log in Start now

AI vindt lekken in minuten: zo blijf je hackers voor

AI finds leaks in websites in hours or minutes, and AI chatbots cause data leaks themselves. What is changing and seven measures you can take now.

Team Netslim 15 Nov 2026 6 min read

Finding a vulnerability in a website used to be work for a specialist with plenty of time. Now AI does a large part of that work, and it is fast. Where an attacker once searched for days or weeks, it now takes hours and sometimes minutes. Not only security experts say so: the Dutch National Cyber Security Centre (NCSC) and the European CERT-EU also warn about it.

For anyone with a website, webshop or customer portal, one thing changes: the time between "there is a vulnerability" and "someone is exploiting it" is almost gone. In this blog you can read what exactly is changing, where AI itself is a cause of data breaches, and what you can do this week.

What is changing: AI searches for vulnerabilities faster and more cheaply

AI can read large amounts of code and search for flaws that people overlooked for years. In May 2026, a researcher from the Dutch security company Hadrian showed the NOS how this works. Using a cheap, publicly available AI model, he analysed the code of a government website, found a security flaw and gained access to protected data, including a database. The cost of that attack: about 10 euros. He stopped as soon as the vulnerability was proven.

The director of the NCSC said that the time to find and exploit vulnerabilities is shrinking from days to hours, and possibly to minutes. Defenders still have a head start, but it is getting smaller.

You also see this internationally:

  • Exploitation before the patch. CERT-EU cites a report that estimates the average time to exploitation of a new vulnerability is now minus seven days. A vulnerability is thus often already being exploited before there is an update. In 2018 there were still 63 days between the two.
  • Vulnerabilities are the number one entry point. For the sixth year in a row, exploiting a vulnerability was the most common way in: 32 percent of the break-ins examined in 2025.
  • Twice as many vulnerabilities found. According to research firm VulnCheck, AI has helped double the number of vulnerabilities found in a year, Reuters reported in October 2026.
  • Immediate consequences. That same week a vulnerability was found in PHP, the language on which WordPress and many other sites run. Government sites in Europe went temporarily offline until they were patched, and large network companies immediately saw attempts to exploit the flaw.

AI as a cause of data breaches in your own business

There is also a quieter side. The Dutch Data Protection Authority received several reports of data breaches caused by employees pasting personal data into an AI chatbot. An employee at a GP practice entered patients' medical data. At a telecom company, a file with customer addresses went into the chatbot.

The problem: the party behind such a chatbot usually stores what you enter. That data then sits on servers of another company, without the customer or patient knowing. According to the regulator, this is often a data breach that you must report.

So even if nobody hacks your website, AI can be the reason customer data ends up in the open.

Why small websites are just as much a target

"We are too small to be interesting" was already wrong, and with AI it is even less true. Attackers do not pick one company to study for days. They let software scan the entire internet for sites with a known weak spot: an outdated plugin, an old PHP version, a forgotten test site. Once one is found, the attack follows automatically.

What they look for on a small site:

  • customer data and orders to sell on or use for blackmail;
  • your server, to send spam or attack other sites;
  • your good name in Google, to secretly place spam pages on it.

The faster the search, the shorter the time a vulnerability can stay open before someone finds it.

What you can do now: seven measures

The good news: the basics stay the same. Only the pace has to go up. An update that could once wait a few weeks now has to be done within days.

  1. Update within days, not weeks. Turn on automatic updates for WordPress, plugins and themes, and keep your PHP version current. Not sure which vulnerabilities are circulating? On our security-updates page we list the new vulnerabilities in PHP, WordPress, Plesk and Linux every morning, with what they mean for you.
  2. Make your site smaller. Remove plugins and themes you do not use, old test sites, old admin accounts and backup files that sit in a public folder. What is not there cannot leak either.
  3. Secure every login with a second step. A stolen password is still the easiest way in. Our guide on setting up two-step verification for WordPress and email explains how to arrange that in a few minutes.
  4. Notice immediately when something changes. An attack you only discover after weeks costs much more. Hackwacht compares your WordPress every day with the original files and shows exactly which file has been changed, so you can restore it with one click.
  5. Stop attacks before they reach your site. A web application firewall blocks known attack patterns, even if you have not yet been able to install an update. All our servers have protection against DDoS attacks as standard.
  6. Keep backups you can really restore. Store them in a different place from your site and test a restore now and then. A backup that does not work is only noticed when it is too late.
  7. Make agreements about AI in the workplace. Agree which AI tools your team may use and which data must never go in: customer data, medical data, contracts, passwords. If it goes wrong anyway, you must report a data breach to the Dutch Data Protection Authority within 72 hours.

Test yourself before an attacker does

The tools attackers use, you can also use yourself. CERT-EU advises organisations to test their own systems with AI, with people checking the outcomes. That way you find the weak spot first yourself.

That is exactly what a pentest does: we try to get in the way an attacker would, and you receive a report with what we found and how to fix it. You can read more about this on our page about cybersecurity, pentests and incident help.

Already hacked, or in doubt? Then first read what to do in the first 24 hours after a hack.

Rules are getting stricter too

It is not only attackers who are speeding up. Since September 2026 the reporting obligation from the Cyber Resilience Act applies in the EU: makers of software must report actively exploited vulnerabilities and serious incidents. Full compliance with that law is mandatory from December 2027. If you build software or apps yourself, this applies to you directly. If you use software from others, it helps you: you hear sooner when something is wrong.

Short summary

  • AI finds vulnerabilities faster and more cheaply. It takes hours, sometimes minutes.
  • Vulnerabilities are often already being exploited before there is an update.
  • AI chatbots are themselves also a source of data breaches when employees enter customer data into them.
  • The basics remain: update quickly, leave few things open, use two-step verification, monitoring and good backups.
  • Want to know how your site stands? Contact us, and we will take a look with you.

Frequently asked questions

Can AI really hack a website by itself?

AI can scan code for flaws and suggest how to exploit them. In demonstrations, cheap models managed to find real vulnerabilities and view data. Usually a human still directs the attack, but the search work is much faster than before.

Is my WordPress site extra vulnerable?

WordPress itself is well secured, but it runs on many sites and has thousands of plugins. An outdated plugin is one of the most commonly used entry points. If you install updates quickly and remove unused plugins, you reduce that risk considerably.

Is it a data breach if an employee pastes customer data into an AI chatbot?

Often, yes. The party behind the chatbot usually stores the input, so the data then sits with another company without the customer knowing. The Dutch Data Protection Authority expects you to report such a breach in most cases, and to inform the data subjects as well.

How quickly should I install an update?

For a vulnerability that is being actively exploited: as soon as possible, preferably the same day. For the rest, within a few days. The time of waiting for weeks until it "comes out" is over.

Sources

Read also

All blogs →

How fast and secure is your site?Free, within a minute

Check my site