HALLUCINATION

Definition 

Hallucination is the AI industry’s term for outputs generated by a large language model that are factually incorrect, fabricated, or unsupported by the model’s training data or provided context — but are presented in confident, fluent, plausible-sounding form. Hallucinations are a fundamental property of how generative language models operate: the models are trained to produce probable continuations of text, not to verify factual claims, and they will produce confident continuations even when the underlying training data is sparse or absent. Hallucination rates vary by model, by topic, by prompt structure, and by whether the model is grounded by retrieval. Reducing hallucinations involves retrieval-augmented generation, fine-tuning for factual accuracy, citation-aware generation, and explicit uncertainty modeling. 

Why it matters for communications 

Hallucination is one of the highest-stakes vocabulary in AI Communications — the term that bridges technical AI behavior into board, legal, and crisis communications. Every AI deployed organization needs accurate communications discipline around hallucinations: what they are, what circumstances elevate or reduce hallucination rates, how the organization manages them, and how legal and regulatory exposure to hallucinations is structured. Hallucination crises — high-profile incidents in which AI-generated content materially misrepresents facts — have become a recurring AI Communications crisis category. 

Related terms Retrieval-Augmented Generation · Alignment · Red Teaming · Evaluations · Grounding 

Related entities OpenAI · Anthropic · Google · Meta · enterprise AI vendors · NIST · academic AI research community 

Primary sources Anthropic, OpenAI, and Google DeepMind safety and evaluation publications · NIST AI Risk Management Framework · academic literature on factuality in language models.