How to Shrink Your Attack Surface Before AI Scanners Find It

Knowing how to shrink your attack surface used to be good hygiene. After this week, it is a priority. On 19 November 2026 the European Commission confirmed that an “AI-generated cybercampaign” had targeted its systems, taking its HR platform Sysper and the MyPMO reimbursement tool offline for around ten days, as reported by Eunews.

According to Brussels Signal, the attack was detected on 22 October and is the first time the EU has publicly disclosed a cyberattack that used AI. The attackers reportedly used a publicly available AI tool to find weaknesses. That is the key point for every European business: AI makes discovery fast and cheap. Whatever you have exposed to the internet, assume it will be found. This guide shows you how to reduce what can be found in the first place, and fix what remains faster than an automated scanner can exploit it.

What You Need Before You Start

Gather these before you begin:

  • A list of your domains and cloud accounts: from your registrar, DNS provider and every AWS, Azure or Google Cloud subscription, including ones finance pays for but IT has never seen.
  • Read access to your firewall and identity provider: Entra ID, Okta or Google Workspace admin.
  • A discovery tool: free options include Shodan, Censys, OWASP Amass and ProjectDiscovery’s subfinder and httpx. Commercial external attack surface management (EASM) tools work too.
  • A named owner: one person accountable for the exposure list, with authority to switch things off.

If you fall under NIS2, much of this also supports your NIS2 risk-management obligations, so document as you go.

Step 1: Discover Everything You Expose to the Internet

You cannot protect what you do not know about. Start by looking at your organisation the way an attacker’s AI tool would.

Enumerate subdomains with Amass or subfinder, then check certificate transparency logs on crt.sh for hostnames you forgot. Search Shodan and Censys for your public IP ranges and company name to see which ports and services are visible. Run httpx across the results to fingerprint web apps, login pages and software versions. Pull public IPs and load balancers directly from each cloud console as well.

Put everything into one inventory: hostname, IP, service, software version, owner and business purpose. Expect surprises. Old staging sites, forgotten VPN appliances, test databases and marketing microsites from 2019 turn up in almost every first scan.

Do not forget third parties. Agencies, payroll providers and SaaS vendors often host pages under your domain or hold API keys into your systems. Add them to the inventory with a contact person and a note on what data they can reach. If a supplier cannot tell you what they expose on your behalf, that is a risk worth raising at the next contract review.

Step 2: Remove or Hide What Does Not Need to Be Public

Every item in your inventory should answer one question: does the public internet genuinely need to reach this? If not, take it off.

  • Decommission dead assets: delete abandoned subdomains and DNS records pointing to cloud resources you no longer own, which also prevents subdomain takeover.
  • Close admin interfaces: RDP, SSH, database ports, Kubernetes dashboards and router admin pages should never be open to the world.
  • Put internal apps behind zero trust access: tools like Cloudflare Access, Zscaler Private Access, Tailscale or Entra Application Proxy let staff reach HR and finance systems without exposing them publicly.
  • Restrict by geography and IP where it makes sense: an internal payroll portal used only in Germany has no reason to accept traffic from everywhere.

Pay special attention to Sysper-style platforms. HR and finance systems hold names, salaries and home addresses. They deserve the strongest isolation you have.

Step 3: Harden What Must Stay Public

Some systems must face the internet: your website, customer portal, email and APIs. For these, raise the cost of every attempt.

Enforce phishing-resistant MFA (FIDO2 keys or passkeys) on every login, starting with admins. Put a web application firewall in front of web apps and APIs, and turn on rate limiting and bot detection, since AI-driven scanning generates exactly the high-volume, patterned traffic these controls are built to catch. Remove version banners and verbose error messages that tell scanners which software you run. Use security headers and disable unused HTTP methods. For APIs, require authentication on every endpoint, including the “internal” ones a developer assumed nobody would find.

Step 4: Patch Exposed Systems on an AI-Speed Clock

If attackers can find a weakness in hours, a monthly patch cycle is too slow for anything internet-facing. Set separate service levels: critical vulnerabilities on exposed systems fixed within 48 to 72 hours, everything else on your normal cycle.

Prioritise using real-world signals, not just CVSS scores. The CISA Known Exploited Vulnerabilities catalog and EPSS scores show which flaws attackers are actually using. ENISA’s EU Vulnerability Database is a useful European reference too. Automate patching where you can, and where you cannot patch quickly, apply a mitigation such as a WAF rule or temporarily taking the service offline.

Step 5: Scan Yourself Continuously, With AI on Your Side

Attack surfaces grow back. New cloud resources, new marketing sites and new SaaS integrations appear every week. Rerun discovery automatically, at least weekly, and alert on anything new or changed.

Use AI-assisted scanning against your own estate before someone else does. Tools like Nuclei with community templates, Burp Suite’s automated scanning and AI-assisted code review can surface weaknesses quickly. Treat the output as a work queue with owners and deadlines, not a report to file. Also rehearse a graceful shutdown: the Commission could switch off several applications as a precaution because it understood its dependencies. Know which systems you can disable quickly, and what breaks when you do.

Common Mistakes to Avoid

Scanning only the IP ranges you remember. Shadow IT, agency-built microsites and old cloud accounts are where the worst exposures hide. Start from DNS, certificates and billing records, not memory.

Treating discovery as a one-off project. A clean inventory in November is out of date by January. Continuous monitoring is the only version that works against automated attackers.

Staying silent with staff. Officials told Euronews they received no formal internal communication about the Commission outage for weeks. When something happens, tell affected people early. It builds trust and helps them spot follow-up phishing.

Key Takeaways

  • Assume discovery is instant: AI tools find exposed systems fast, so reduce what is exposed.
  • Inventory from the outside in: DNS, certificate logs, Shodan and cloud consoles reveal what attackers see.
  • Hide internal apps: HR, finance and admin tools belong behind zero trust access, not on the open internet.
  • Patch exposed systems in days: use KEV and EPSS to decide what goes first.
  • Keep scanning: weekly automated discovery catches new exposures before attackers do.

Need Expert Help?

If this feels like a lot to manage alone, TecniForge can handle the heavy lifting. Our team specializes in custom software development and AI integration. Get in touch with our experts.

Also read: EU Commission Cyberattack: 7 Lessons From the First Confirmed AI-Driven Attack on Brussels, our earlier coverage on why this matters today.

Try this today: search your company name on Shodan and look up your domain on crt.sh. If anything appears that you did not expect, an AI scanner has probably already seen it too.


Discover more from TecniForge

Subscribe to get the latest posts sent to your email.