Chapter 09 · Sustainability & AISustainability & AI

AI confabulation

Meaning statusEstablishedSource recordDirect document linkedWhy these are different

Definition

A false or unsupported output generated by an AI system and presented as if it were reliable; commonly called an AI hallucination.

References

NISTAI RMF Generative AI Profile (NIST AI 600-1, 2024)

This source provides part of the technical or institutional basis for the definition.

EUR-LexDirective (EU) 2024/825 — Empowering consumers for the green transition

This source supports the explanation of how the term is applied, measured or governed in practice.

NeurIPS 2020 / arXivLewis et al. — Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks

This source supports the wider sustainability significance and context described in the entry.

Overview

What it means

NIST uses confabulation as the formal term for confidently presented false or erroneous content, while hallucination remains the familiar colloquial label. It occurs because generative models produce statistically plausible outputs rather than independently verifying each claim. Grounding techniques such as retrieval-augmented generation can reduce, but not eliminate, the risk.

How it is used

In sustainability practice the risk is concrete: invented figures, misattributed standards, fabricated citations in AI-drafted disclosures or analyses. Any AI-assisted reporting workflow needs source-checking before publication.

Why it matters

Environmental claims face tightening substantiation rules; the EU's Empowering Consumers Directive bans unsupported generic claims from September 2026. An AI system that invents green credentials creates legal exposure, not just embarrassment.