What AI does to liability
A client cannot recheck an opinion, an audit or a diagnosis. Work like that becomes credible only because the provider is liable for mistakes. Artificial intelligence attacks that mechanism.
What happens
Provider and checker reach for the same model. Once the check turns up nothing the provider would not have had anyway, a liability pledge says nothing about the quality of the work.
Who it hits
Occupations whose work the client cannot judge: audit, medicine, law, expert opinion, technical certification. First where the fee is small and the possible loss is large.
What helps
Three conditions have to hold at once: the mistake must be provable, the law must let the loss reach the provider, and the provider must be able to carry it. If one is missing, no certificate makes up for it. Which occupations still meet all three is what the figures below show.
The survival ridge
A buyer cannot tell whether an opinion, an audit or a diagnosis is sound — they rely on the provider answering for mistakes. Each point is one occupation in one legal system, the surface the line where that liability stops working. The drawing turns by itself and you can take over with the mouse — the full page explains why the line falls where it does.
- still holds
- no longer holds
Where a liability pledge still holds
A provider who answers for failure stands apart from one who does not — but only where the law lets the loss reach the pledge and the stakes sit in a workable range. Five professions, three legal systems.
| Germany | United Kingdom | USA | |
|---|---|---|---|
| Software development | |||
| Legal advice | |||
| Radiology | |||
| Audit & assurance | |||
| Social work |
- Window open — the pledge works for most engagements
- Narrow window — only applies for a part of the market
- Closed — the pledge cannot separate
Germany opens the widest window: courts let the loss reach the pledge and void caps set too low.
In the United States it is shut: damages are capped at the actual loss, so the pledge cannot promise more than what already happened.
Social work stays closed everywhere: the harm is decoupled from the fee, so no pledge can scale with it.
Five professions × three legal systems, from the Institutional Window model — an illustration built from published error-rate and liability data, not a statement about any specific contract. the paper →
Advisory-led, research-backed. Advice you can recheck.
Aegis Compliance and Strategies OÜ advises on compliance and strategy. The advice carries; the research backs it. Every recommendation points to a disclosed, machine-checked paper — and this site contacts no third party while doing so.
Event-driven Advisory
Compliance and strategy, ordered by your situation. Four occasions, each with a named basis and condition.
→ AdvisoryTwo programmes, one bridge
Six disclosed papers. Law as a parameter versus law as a state variable.
→ ResearchRegistry, gate, open points
Every number on these pages comes from a versioned registry and is checked after every build.
→ Method