Fake invoices are one of those problems that sound boring until you see the numbers. Sweden's criminal economy generates approximately SEK 352 billion annually, about 5.5% of the country's GDP. Across the European Economic Area, payment fraud losses reached EUR 4.2 billion in 2024, up 17% year-on-year. Credit transfers, the primary channel for fraudulent invoices, accounted for EUR 2.5 billion of those losses.

Swedish cyber intelligence firm Njordium Cyber Group AB just launched a product it hopes will put a dent in those figures. Its new AI Fraud Detection Module, integrated into the company's recently released Vendor Management System, uses self-learning artificial intelligence to detect fake invoices, phantom services, and inflated pricing in real time.

The timing is deliberate. As organizations digitize procurement and vendor management, the surface area for invoice fraud expands. And the fraudsters are getting better.

The Expert Group on Public Economics (ESO) published a report earlier this year, Svarta siffror (2026:1), that laid out the scale of Sweden's criminal economy in uncomfortable detail. The shadow economy alone accounts for about SEK 224 billion per year, driven by undeclared labor, tax evasion, and fraudulent commercial activity. A significant portion of that flows through doctored invoices, a method that exploits the trust businesses place in their vendor relationships. The EBA and ECB reported that 85% of credit transfer fraud losses fall directly on victim organizations, with little to no route to recovery.

That's the core of the problem. When a fake invoice gets paid, the money is usually gone. Recovery rates are dismal. Prevention is the only realistic strategy, and prevention requires catching fraud before payment, not after.

Most existing fraud detection systems are rule-based. They flag invoices that match known patterns: duplicate amounts, suspicious vendor names, mismatched bank details. The problem is that sophisticated fraudsters know the rules. They adjust. A rule-based system is always fighting the last fraud, not the next one.

Njordium's module takes a different approach. The AI engine learns continuously from user feedback and transaction patterns, adapting to new fraud techniques as they emerge. When an accounts payable team flags or approves invoices, the system gets smarter. When new fraud patterns appear across Njordium's client base, the model updates.

The company is also emphasizing EU AI Act compliance, which is becoming a competitive differentiator rather than just a regulatory checkbox. As the AI Act's requirements roll into enforcement, vendors that can demonstrate transparency and auditability in their AI systems will have an easier time selling to regulated enterprises.

Njordium started as a cyber intelligence and governance consultancy. The Vendor Management System it launched earlier in March was its first product move, creating a centralized platform for enterprise vendor oversight and financial risk management. Adding AI fraud detection on top of that platform is the kind of product expansion that looks obvious in retrospect but required building the underlying infrastructure first.

The company is positioned squarely at the intersection of cybersecurity and procurement, two domains that have historically operated in silos. Procurement teams manage vendor relationships. Security teams manage threat detection. Njordium's bet is that combining both into a single platform creates value that neither function can achieve alone.

Any organization with a significant accounts payable operation is a potential customer. But the most immediate targets are mid-to-large enterprises in regulated sectors, companies that face both regulatory pressure on vendor due diligence and material financial exposure to invoice fraud.

The Swedish public sector, where procurement volumes are high and fraud oversight has historically been reactive, is another natural market. The ESO report's findings are likely to accelerate government interest in exactly this kind of technology.

Njordium hasn't disclosed pricing or customer targets for the module. What it has disclosed is that the module is live and operational. In a market where most AI fraud detection is still in pilot, shipping product is its own competitive advantage. The question now is whether enterprises adopt it fast enough to make a measurable dent in that EUR 4.2 billion annual loss figure.