Slovenia's insurance watchdog will test insurance AI in a sandbox before it reaches you

Think about the last time you filed a car insurance claim. In much of Europe, an algorithm probably helped decide how quickly it was settled — and whether it was paid. Now the people who supervise those algorithms in Slovenia have chosen a tool to supervise them. On August 6, 2026, AZN announced it will use the AI governance platform built by Modulos AG — not only for its own internal AI systems, but as a regulatory sandbox for the insurers and technology vendors it oversees, as EIN News reported.

Why a supervisor needs its own AI platform

Insurance companies in Slovenia, as elsewhere, increasingly rely on machine learning for underwriting, fraud detection and claims handling. A system that flags suspicious claims can save money; a system that silently rejects borderline cases can create serious legal trouble for both the insurer and the customer.

The EU AI Act treats many of these uses as high-risk. That means they must meet strict requirements for data quality, transparency, human oversight and documentation — and prove it through a conformity assessment. For a regulator, auditing such systems with clipboards and spreadsheets is no longer realistic. You need software to review software.

That is what AZN is now setting up. The Modulos platform gives a supervisor a complete inventory of AI systems, a way to assess their risk under the AI Act, and tools to monitor them over time — rather than only at the moment of approval.

The timing matters: the AI Act deadline just moved

Here the story gets genuinely European. In July 2026, the EU's "Digital Omnibus on AI" — officially Regulation (EU) 2026/1744 — deferred the conformity obligations for standalone high-risk AI systems such as recruitment tools and credit scoring to December 2, 2027. Insurance-related systems that fall under Annex III are part of that group.

The extra time is not a free pass. Since August 2, 2026, national authorities have had full enforcement powers over general-purpose AI and transparency rules under Article 50 — including fines for failing to label AI-generated content or to tell people they are talking to a chatbot. Financial supervision adds its own layers: DORA (digital operational resilience for financial firms) and NIS 2 (cybersecurity) already apply to insurers and their technology providers.

Slovenia is therefore preparing early. Its sandbox is designed to test AI systems in a controlled environment, with regulatory oversight, before full market deployment. For a small insurer, that can mean the difference between a costly, opaque compliance project and a guided process that actually leads to approval. You can follow our ongoing coverage of the AI Act's practical impact in the ai-jarvis.eu magazine.

What the platform actually does — in plain language

Modulos' software is not a magic scanner. It is closer to a hygiene system for AI operations:

• Inventory: it maps every AI system an organisation runs, including the ones nobody remembers to report.
• Risk scoring: it classifies each system according to the AI Act's categories and flags the ones that will need a conformity assessment.
• Documentation: it automatically generates the technical records a regulator will ask for during an audit.
• Monitoring: it keeps watching systems after deployment, alerting when a model starts behaving differently.

For AZN, the same tool serves two purposes: governing its own internal AI and reviewing submissions from the market. That dual role is what makes a regulator's sandbox credible — the supervisor is using the very discipline it asks of others.

A young European company on a growing market

Modulos is a European player, not a Silicon Valley giant. The Zurich-based company has raised CHF 16.4 million in total, including a CHF 8.7 million pre-Series A round (about €9.3 million). In August 2026, it was also named in the inaugural Gartner Magic Quadrant for AI Governance Platforms.

The symbolism matters. AI governance is becoming a market of its own — with specialist vendors, analyst ratings and national authorities as clients. When a sovereign regulator picks a European supplier for this, it is also a practical answer to the question of where European "digital sovereignty" lives: not only in grand data centres, but in the quiet bureaucratic tools that make supervision possible.

What this changes for an ordinary customer

For most people, the visible effect will be indirect. You will not see the sandbox. But when a claim is declined by an automated system in Slovenia, there is a greater chance that the system has paperwork proving why — and that a human can trace the decision. The EU's transparency rules, enforceable since August 2, 2026, already require clear disclosure when a customer is talking to a chatbot rather than a person.

There is also a risk worth naming. If every national regulator chooses a different governance platform, Europe ends up with a patchwork of tools and methods that other countries' vendors will not understand. Brussels has not yet mandated a single format for AI documentation. A useful next step would be common standards that let a Slovenian sandbox certificate be recognised in Vienna or Warsaw.

For now, the practical advice for European businesses is simple: the countdown to December 2, 2027 has begun, and it is shorter than it looks. Start with an honest inventory of your AI systems, classify them, and find out which ones are high-risk. The regulator in Ljubljana just showed that this can be done with a commercial tool — and tested before the law demands it.