← ALL NEWS

INGENIØREN · 12 Nov 2026 · DENMARK

From 14 days to 20 seconds: How TDC Erhverv uses AI to detect threats in real time

TDC Erhverv, which manages internet traffic for roughly half of Denmark's businesses, has built an AI security system that detects threats in its customers' network traffic within seconds. Martin Nissen, head of the Foundry (the company's in-house R&D and innovation lab), says the most recent alerts went out less than 20 seconds after the incident. By comparison, Mandiant's latest report shows the average time for a Danish company to detect a breach has risen from 10 to 14 days over three years, with small and medium-sized enterprises pulling the figure up.

About nine billion signals pass through the network each day, a volume Nissen says could not be analysed without AI. Three years ago the traffic data was used only to check that customers received the service they paid for. The company now focuses on "early kill chain indicators", meaning early signs of malicious activity, so attackers can be caught as early as possible.

The system is a three-stage analysis engine working on metadata only. The company says it never inspects the content of customers' traffic. Parameters include data volumes, packets, timestamps and ports, enriched with external data. The first stage filters out noise such as streaming services and notifications, while checking all traffic against known threat indicators from global databases. The second stage uses embeddings and unsupervised learning rather than hand-written rules, which lets it spot unexplained "islands" in the traffic. These go to a locally run language model trained on malware behaviour, published security reports and open source intelligence. That model can then actively probe the suspicious device to determine its purpose. If malware is found, customers can be notified almost immediately.

Unsupervised learning matters because manually retraining the model would leave it constantly behind in a fast-changing threat landscape. The solution runs on-premises at TDC Erhverv. Nissen notes that the longer attackers operate undetected, the more severe the damage, so fast detection directly limits the consequences for businesses.

Read the original ↗