Resources

Research and frameworks for governed, insurable AI.

Our working material on the market, the law and how we measure AI risk. Available on request while we prepare public versions.

Request a demo Risk framework

Market report · October 2026

AI governance and liability insurance in Germany and the EU.

Market structure, carrier appetite, competitors and entry barriers.

Request the market report

Carriers: No German carrier offers a dedicated AI-liability product, and standard liability wordings are largely silent on AI.
Governance vendors: European AI governance platforms produce documentation and evidence, but none attach insurance.
Insurance entrants: Insurance-backed AI assurance has started in the US and UK, without an EU entity or EU standards mapping.
Standards: EN 18286 gives carriers and companies a shared definition of governed AI for the first time.

Regulatory briefing · October 2026

The Product Liability Act 2026 and the AI Act Omnibus.

What changes on 9 December 2026 and which documentation to have ready.

Request the briefing

Scope: Software and AI become products under strict liability, including updates and machine-learning behaviour after sale.
Disclosure: Courts can order a defendant to disclose relevant evidence, including technical documentation.
Presumptions: A defect can be presumed when evidence is withheld or the claimant faces excessive difficulty proving it.
AI Act timing: The Omnibus moves high-risk obligations to 2 December 2027 (Annex III) and 2 August 2028 (Annex I).
What to prepare: An inventory, logs, evaluation records and incident history for every AI system you put on the market.

Risk framework

Use, Build, Ship.

The control codes behind the AssurAI risk score. Use covers AI your people use, Build covers AI you develop, Ship covers AI you put in front of customers.

Request the full framework

U-01Unknown model in use: A model is in use with no owner and no record of what it reaches.
U-03Unapproved connector: A connector reaches company systems without approval.
U-07Personal data in prompts: Customer or personal data is sent to a model without a lawful basis or control.
B-02Unversioned prompt or skill: A prompt or skill runs in production without version control.
B-05Configuration drift: Teams run different configurations of the same AI system.
S-02Unrecorded incident: An AI incident is not recorded or escalated.
S-04No post-market monitoring: A live AI system has no monitoring of its behaviour after release.
S-07Incomplete documentation: Technical documentation is incomplete at the time a court or regulator asks for it.