Article

Who is responsible when an AI responds without legitimacy?

An AI system does not carry responsibility. Yet its responses are increasingly used as if they were reliable, actionable, and enforceable. Responsibility therefore follows the governance chain, not the model alone.

EN FR
CollectionArticle
TypeArticle
Categoryrisque interpretatif
Published2026-01-27
Updated2026-03-15
Reading time4 min

This article closes the loop. AI bears no responsibility. Yet its responses are increasingly used as if they were reliable, actionable, and enforceable. When a response becomes contestable, the question immediately surfaces: “who is responsible?”. The answer is rarely comfortable, because interpretive risk is not a tool problem. It is a responsibility chain problem.

The false debate: blaming the model

Blaming “AI” is a way to mask the real subject. The model produces a response within a usage context defined by an organization, in a channel, with objectives (respond quickly, reduce escalations, automate). What the model then does, it does under implicit constraint: produce a response. The problem is not that AI “gets it wrong”. The problem is that it responds without interpretive legitimacy. See symptom requalification: /blog/interpretive-risk/hallucination-absent-interpretive-legitimacy/.

Responsibility never disappears

In a real context, responsibility shifts toward those who determine:

  • what AI is authorized to say (perimeter);
  • which sources are authoritative (hierarchy);
  • how contradictions are handled (explainable arbitration or refusal);
  • what the system does when information is missing (indeterminacy or non-response);
  • who assumes use of produced responses (in a given channel).

In other words: responsibility follows governance, even when it is implicit.

The three places where responsibility crystallizes

1) Publication and attribution

Once a response is published on an institutional surface (site, chatbot, support, communication), it is perceived as attributable. The organization assumes consequences, even if the response was automatically generated. See the public communication case: /blog/interpretive-risk/public-communication-ai-official-position/.

2) Actionable use

When a response influences a decision (HR, legal, operational), responsibility shifts toward the act of use. The problem is no longer generation, but employing the output as a decision basis. See the HR case: /blog/interpretive-risk/hr-when-ai-inference-becomes-a-discrimination-risk/.

3) Contestation and enforceability

Contestation reveals the central question: is the response defensible without fiction? If the justification chain is not reconstructible, responsibility expresses as exposure: legal, economic, reputational. See the role of source hierarchy: /blog/interpretive-risk/source-hierarchy-enforceability/.

Why enforceability forces responsibility

An enforceable response is one that can be defended. Therefore, an enforceable response implies that an organization can explain:

  • which sources it relies on;
  • which perimeter it authorizes;
  • which exclusions prohibit certain inferences;
  • how contradictions were handled;
  • why non-response was not chosen.

Without this structure, responsibility exists anyway, but in its most costly form: uncontrolled exposure.

The key point: non-response is a responsibility mechanism

When information is missing, when sources contradict, or when the question crosses a commitment boundary, forcing a response amounts to manufacturing a liability. Legitimate non-response is a way to preserve contestability and avoid unauthorized inference. See informational silence: /blog/interpretive-risk/informational-silence-legitimate-non-response/.

What interpretive governance changes

Interpretive governance does not “shift” responsibility toward AI. It makes it explicit by governing response conditions:

This framework does not eliminate error. It reduces the space where error becomes indefensible.

Anchor

Responsibility does not disappear with AI. It simply becomes harder to assume when response conditions are not governed. Making a response enforceable means making responsibility explainable, bounded, and defensible, rather than suffered.