Space Regulatory Glossary AI & Data Science

hallucination

/həˌluːsɪˈneɪʃən/ noun

1. An AI model output that is factually incorrect, fabricated, or unsupported by the source data, presented with the same apparent confidence as accurate information. Hallucinations range from invented citations to subtly wrong numerical values.
2. In regulatory contexts, hallucinated docket numbers, fabricated rule citations, or incorrect filing dates can lead to compliance failures. Mitigation strategies include RAG architectures, confidence scoring, and verification against authoritative databases.

Read: The Trust Problem — When AI Hallucinates Regulations

Related terms

large language model A neural network trained on large text corpora that generates, summarizes, and reasons about natural language. Models… retrieval-augmented generation An AI architecture that grounds a language model's output in retrieved source documents rather than relying solely on… confidence scoring A method of attaching a numerical reliability estimate to an AI system's output, indicating how certain the system is…

Browse all terms in the Space Regulatory Glossary →