German web host breached through AI assistant connector

Mittwald, a German web hosting company, disclosed on 9 November 2026 that hackers had breached its systems through a tool designed to let customers connect AI assistants to their accounts. According to a report by The Record, the company said the attackers were allegedly assisted by AI.

The entry point was the connector itself — the feature that allows a customer's AI assistant to reach into their hosting account and act on it. That is the part of the stack that security teams have been warning about for some time: a connector is, by design, an authenticated path into a live account, and it exists precisely so that software rather than a human can use it.

The disclosure is thin on detail. The available reporting does not say when the intrusion began, how long the attackers were inside, how many customers were affected, or what data or systems they reached. Nor does it say what specific weakness was abused, whether the flaw was in the connector's design, its authentication, or in how customers had configured it. The company has not, in the reporting seen so far, set out how it detected the breach.

The claim that AI assisted the attackers is also unqualified in the material available. It is not clear from the disclosure what role AI is said to have played, whether in reconnaissance, in crafting the intrusion, or in driving the connector once access was obtained. Readers should treat the "AI-aided" framing as the company's characterisation rather than an independently verified finding. No attribution to a named group has been reported.

What is confirmed is the shape of the incident: a hosting provider, a customer-facing AI integration, and a breach that ran through it. For other providers shipping similar features, the practical questions are the familiar ones. Which actions can a connected assistant take without a human in the loop? Are those actions logged separately from ordinary customer activity, so that an assistant behaving strangely can be told apart from a customer working late? Can a connector's permissions be revoked quickly across all accounts if one is abused?

The Agentic Times has not independently verified the company's account. We will update this report if Mittwald publishes a fuller technical post-mortem, including the root cause, the timeline and the number of accounts involved.