English

The EU drops its AI liability rules. So who pays when AI gets it wrong?

The Commission withdrew the AI Liability Directive on 11 February 2025. The question it was meant to answer did not go away.

On 11 February 2025, in its work programme for the year, the European Commission announced it would withdraw the proposed AI Liability Directive, citing no foreseeable agreement and the broader push for regulatory simplification.

That directive was meant to make it easier for someone harmed by an AI system to bring a claim — by easing the burden of proving that a system caused the damage. Without it, ordinary national liability law applies.

Why this matters more to an SME than it sounds

Because you are more likely to be the one holding the liability than the one claiming it.

Consider the realistic scenario. An AI feature in your quotation software mis-states a delivery period. The customer relies on it. You are late, they claim damages. Who is responsible?

The answer today, in Switzerland and in most of Europe: you are. You sent the quotation. Your software vendor’s contract almost certainly limits their liability to the licence fee, and the model provider behind them has no relationship with you at all.

The withdrawn directive would not have changed that chain much. It is worth understanding precisely because nothing is coming to change it.

Three practical consequences

  • Read the liability clause in AI-touching software contracts. Not to negotiate it — you probably cannot — but to know what it says before you build a process on top of it.
  • Keep the human confirmation step where money or commitments are involved. It is not only good practice. It is the thing that makes the error yours to catch rather than yours to explain.
  • Ask your insurer once. Whether your professional liability cover responds to an error produced with AI assistance. Most policies do not mention it, which means the answer depends on wording rather than intent. Better to know now.

The through-line

Regulatory attention is moving from restricting AI to encouraging it. That does not reduce your exposure — it leaves it exactly where it has always been, with the company that signed the document.

Which is another way of saying what this blog keeps saying: the tool drafts, a person decides. That principle is not compliance theatre. It is where your liability actually sits.

← Back to all posts