NEWS

Alleged AI-Aided Cyberattack Exploits Assistant Interface to Breach German Web Host

German host Mittwald says an attacker abused its MCP server, which connects AI assistants to customer accounts, to compromise 56 customer projects.

Dylan H.

News Desk

November 9, 2026
2 min read
Alleged AI-Aided Cyberattack Exploits Assistant Interface to Breach German Web Host

Overview

Mittwald CM Service, a German web hosting provider for agencies, has disclosed a breach in which an attacker, allegedly aided by AI tools, abused the company's Model Context Protocol (MCP) server. The server lets customers connect AI assistants to mStudio, Mittwald's platform for managing websites and hosting.

What Happened

  • The attacker first probed the MCP flaw in October, during a broader rush to patch the PHP2Shell vulnerability.
  • The intrusion went unnoticed for eight days.
  • The attacker moved quickly and used disguised access attempts and diversionary tactics.
  • Mittwald says its logs point to AI tooling because steps happened faster than a human could manage. It has not published other evidence or explained how it distinguished AI tools from conventional scripts, so the AI claim is unproven.

56 customer projects were compromised. Mittwald has not said what data was accessed or taken.

It is unclear whether the flaw lay in Mittwald's own code or in open-source MCP libraries.

Response

Mittwald disclosed the breach in a blog post on Saturday. On the preceding Friday it reported no further attacker activity after forensic reviews and repeated scans. External forensic experts are assisting, relevant authorities have been informed, and a detailed account is promised the following week once analysis and recovery finish.

Why It Matters

MCP servers give AI assistants delegated access to real accounts and infrastructure, so they widen the attack surface in the same way as any new authenticated API. A flaw in one can hand an attacker the same reach as a legitimate assistant.

Recommendations

  1. Mittwald customers: watch for the company's follow-up report, and review the 56-project notification process for your account.
  2. Rotate credentials, API tokens, and SSH keys tied to affected hosting projects.
  3. Inventory any MCP or AI-assistant integrations that can act on your accounts, and apply least-privilege scopes.
  4. Log and alert on MCP server activity, including unusually rapid sequences of API calls.
  5. Keep MCP libraries and servers patched, and do not expose them to the internet without strong authentication.