Guidance · Security and AI

Your webshop is only one way in.

Attacks are now run by swarms of automated AI agents that scan, probe and break in around the clock. Nobody has to pick you as a target. They're after your data, not just your uptime. Here's what to secure now, and how to use AI yourself without breaking the EU AI Act.

Updated 21 November 2026

01 · The threat

Attacks are more sophisticated now

AI swarms

AI agents orchestrate other agents to attack your systems from all sides, all working towards the same goal in parallel. That's why normal defences aren't good enough anymore. And they attack indiscriminately.

A fraction of a second

AI is discovering fatal flaws in the most popular systems, flaws that used to be impossible to exploit. Attackers now use them to bypass nearly any form of protection you have, which makes auditing your infrastructure a business-critical matter.

It happens constantly

We see attacks take place nearly every hour of every day, and nobody is exempt. This year alone, attackers slipped malware onto the websites of IKEA, Zoom and Salesforce through one supplier's script, and hit OpenAI and Vercel through compromised open-source tools.

02 · The attack surface

Your whole setup is the target

Locking down the webshop is a good start. But attackers go for the weakest door, and that's rarely the storefront.

Domains, DNS and cloud accounts

Whoever controls your domain controls your email and your site. Lock down the registrar and the cloud console.

Email and office accounts

AI-written phishing reads perfectly in any language, and one stolen password opens mail, files and every system that trusts it.

Forgotten systems

Old test servers, copies of the shop with real customer data, a subdomain nobody remembers. Attackers find them first.

ERP, accounting and stock

They hold your customers, prices and supplier terms, and they're wired into everything else.

Staff laptops and phones

One infected laptop with a saved login is a way into everything that person can reach.

Integrations and API keys

Every link between two systems is a key. Old keys are rarely changed and often have far more access than they need.

Backups

Ransomware goes for your backups first. If they sit on the same network, they go down with everything else.

Suppliers and agencies

Plugins, extensions and partners with admin access: their weak spot becomes yours.

03 · The real cost

Downtime is the cheap part

A shop that's down for a day hurts. Having your customers' data stolen can bring your entire business to the ground.

Stolen data

Names, addresses, order histories, sometimes logins and payment details, sold on or used for fraud in your name.

GDPR

A personal data breach has to be reported to the data protection authority within 72 hours, and often to every affected customer too. Fines go up to €20 million or 4% of global annual turnover.

NIS2

NIS2 applies to many companies, and it's been in force in Denmark since 1 July 2025. Most webshops aren't covered themselves, but if you supply a company that is, its security demands reach you through the contract.

Extortion and fraud

Stolen data gets used for blackmail, fake invoices to your customers and suppliers, and convincing scams in your name.

Trust

Customers may revisit a slow site. But they won't revisit one that lost their data to hackers.

04 · What to do now

Start here

None of this needs a big programme. It needs someone to look at the whole picture and get the basics right.

  1. 01

    Map it all

    Every system, account, integration and supplier, and who has access to what. You can't protect what you don't know you have.

  2. 02

    Two-factor everywhere

    On email, admin panels, cloud, banking and the domain registrar. No shared logins, and close accounts the day people leave.

  3. 03

    Patch immediately

    Webshop, extensions, servers and office software. Automate what you can, and watch for critical fixes.

  4. 04

    Backups you've tested

    Kept apart from the network, and actually restored at least once. A backup you've never restored is a hope, not a plan.

  5. 05

    Least access

    Integrations, API keys and staff get only what they need. Change keys regularly and delete the ones nobody uses.

  6. 06

    Watch and log

    Detect a break-in instantly, not later, because every second matters. And defend with your own AI swarm.

  7. 07

    Have a plan

    Who's your first point of contact? Who calls who? When do lawyers get involved, and when do the authorities? Your first point of contact should have the answer to all of it.

  8. 08

    Test it

    Have someone try to break in before an attacker does, and fix what they find.

05 · AI and the law

Using AI? The EU AI Act applies

The EU AI Act (Regulation (EU) 2024/1689) is in force and applies in stages. Most businesses are what the Act calls deployers: you use AI, you don't build it. That still comes with obligations.

What applies when

  1. Since 2 February 2025

    Banned uses are illegal, such as manipulating people, social scoring and scraping faces from the internet. And you're expected to make sure your people understand the AI they use.

  2. Since 2 August 2025

    Rules for the providers of general-purpose AI models, the big language models, apply.

  3. Since 2 August 2026

    Transparency: people have to be told when they're talking to an AI, and realistic AI-generated images, audio or video (deepfakes) have to be disclosed.

  4. From 2 December 2027

    Strict rules for high-risk uses, such as AI in recruitment, credit scoring or education. Moved from August 2026 by an amendment adopted in 2026.

Fines go up to €35 million or 7% of global annual turnover for the worst breaches.

Read the AI Act on EUR-Lex

Be careful with

Customer data in AI tools

Putting personal data into an outside AI service is a GDPR question: you need a data processing agreement, and to know where the data is stored and whether it's used for training.

Secrets in chatbots

Contracts, passwords, price lists and customer lists don't belong in a public chatbot.

AI that acts on its own

AI agents with access to your systems can be tricked by hidden instructions in a product text, an email or a web page (prompt injection). Give them as little access as possible, and log what they do.

Decisions about people

AI in hiring or credit checks counts as high-risk under the AI Act. Anywhere AI decides something about a person, a human should check it.

AI you didn't know you had

Many tools you already use have AI built in. Find out which ones, and what data they send where.

A simple register

Which AI tools, for what, with which data, and who's responsible. It's your first answer when a customer, an auditor or an authority asks.

Not sure where you stand?

We look at the whole setup: the store, the servers, the integrations and the AI you use, and tell you straight what to fix first.

This page is general guidance, not legal advice.