Where AI actually gets hard: when it has to hold up

Hey all, Bernhard from Hamburg.

Almost 30 years as an entrepreneur (strategy, asset management), former CEO of an IT enterprise company. Today I build myself, daily with Claude Code, Codex, Lovable and Supabase.

My focus is the part almost every AI builder skips: the intersection of AI, law and liability. Everyone ships automations. The real question comes right after: does it hold up in court, who's liable, is it auditable? That's where most people go quiet.

My approach: compliance not as a slide deck, but in code. Deterministic, versioned, signed, provable. Built, not claimed. A legal obligation becomes a running system, not a concept in a drawer.

Two paths, concretely:

Suite: ready-made compliance products. DeeplySecure (cybersecurity audit + EU NIS2), NoviGuard (early-warning under German StaRUG, protecting directors from personal liability).
Build: custom AI systems, same approach. An AI audit checks where AI actually holds up, a blueprint defines roles, data and responsibilities, then the live system goes into operation. Fixed price, handover included.

Behind it sits a scientific advisory board across law, controlling, insurance and cybersecurity that signs off the methodology before any code is written. The kind of depth a single builder doesn't have.

Why I'm here: this community builds instead of talking. If you're shipping AI for clients and you hit compliance, liability or auditability, come talk to me. That's my machine room. 👊

4 comments

Where AI actually gets hard: when it has to hold up