How AI is changing website security, and what a small business can do
Attackers use AI to automate scanning and scam texts, and AI chatbots on your site add new risks. What a small company can do and how we approach AI integrations.
AI is changing everyone's work, including that of people trying to get into other people's websites. For the owner of a small site or e-shop, two things follow: attacks are cheaper and easier to scale, and new AI features on a site bring risks of their own. Here is an overview without the scare tactics.
Why small sites are in the crosshairs too
It used to be possible to rely on a small site interesting nobody. That was never entirely true, and it is even less true today. Attackers rarely pick a specific target. They run automated scanners that go through a huge number of sites looking for known weaknesses: an outdated plugin, a weak password, an unprotected admin. AI makes their job easier because it speeds up writing tools and can produce convincing scam emails and messages without typos, in Czech, tailored to a particular company.
The consequence is simple: it does not matter whether a site is small. It matters whether it has known holes.
New risks: AI directly on your site
If you use an AI chatbot, a form assistant or automatic replies on your site, there are risks a classic website did not have:
- Prompt injection. A visitor types an instruction into the chat to try to override the chatbot's instructions, for instance to reveal internal information or do things it should not.
- Data leaks. If the AI has access to orders, customers or internal documents, a bad setup can hand them to someone they do not belong to.
- Cost abuse. A publicly available chatbot is paid for per use. Without limits someone can flood it and you get an unexpected bill.
- Wrong or invented answers. AI can confidently say something untrue, for example about prices, terms or availability.
What a small business can do
None of this needs a big budget:
- Update. Site core, plugins, themes, PHP. Still the most effective measure.
- Turn on two-factor authentication for the admin, hosting, email and domain.
- Give the least privilege necessary. Every user and every tool, AI included, should have access only to what it needs. An AI assistant allowed to read only public product information cannot reveal the customer database.
- Don't paste customer data into public AI tools. Names, addresses, orders and internal documents do not belong in an ordinary AI chat unless you know how the service handles them.
- Review what AI integrations can access. Which services are connected, what data they see and who can change them. Cancel unused access.
- Keep backups that can be restored, and try it. It is insurance when anything goes wrong.
- Set limits. For public AI features, limit the number of requests, input length and monthly budget.
How we approach AI integrations
When we add AI to a site, we stick to a few principles. The AI gets access only to the data it really needs. Sensitive operations, such as changing an order or issuing a refund, we do not leave to its answer but to a verified step in the application. We add limits so the budget cannot be abused. And we do not treat the output as certain: where accuracy matters, a person or fixed rules check it.
No system with AI is bulletproof. But it can be built so that even after a mistake or misuse the damage stays small.
How such a check looks for a website as a whole is described in Website and e-shop security audit. If you already suspect something is going on, read How to tell your website has been compromised.
Want to check how your site is doing, or think an AI feature through so it is safe? See our security service or write to us.
More articles
Website and e-shop security audit: what we check and what you get
Logins, updates, forms, payments, HTTPS and backups. What a website security audit covers, what the final report looks like and what an audit does not guarantee.
GDPR and cookies on your website and e-shop: what we look at in a check
Cookie banner, third-party scripts, forms, privacy policy and e-shop terms of sale. What we check on the technical side and what a lawyer has to assess.