Hallucination

Definition

Hallucination is the failure mode where a large language model generates plausible-sounding but factually incorrect information with apparent confidence. Studies show unmitigated LLMs hallucinate on 15-30% of factual queries, making hallucination mitigation a mandatory engineering requirement -- not a nice-to-have -- for any production AI system that surfaces facts.

LLMs predict the next most-likely token. When a model does not know the answer, it still generates fluent text -- and that text is often wrong. The danger is that hallucinated outputs look identical to correct ones. Users cannot distinguish without an independent check.

Hallucination mitigation strategies

RAG -- ground every response in retrieved source documents; cite the source
Output validation -- run a second model pass or rule check against known-good data
Constrained output -- limit the model to selecting from a verified list rather than free generation
Human-in-the-loop -- require human review for high-stakes outputs (medical, legal, financial)

Hallucination and liability

In regulated industries (healthcare, legal, GovCon), AI systems that surface unverified facts create real liability. Every enterprise AI deployment must have an explicit hallucination mitigation plan documented before go-live.

Related terms

RAG (Retrieval-Augmented Generation)

Retrieval-augmented generation (RAG) is an AI architecture that supplements a large language model's static training knowledge with real-time retrieval from a private or external knowledge base. RAG reduces hallucinations by grounding LLM responses in verified source documents, making it the standard pattern for enterprise AI assistants built on proprietary data.

LLM (Large Language Model)

A large language model (LLM) is a deep-learning model trained on billions of text tokens to predict and generate human-readable language. LLMs such as GPT-4, Claude, and Gemini power chatbots, document summarization, code generation, and AI workflow automation -- and serve as the reasoning engine inside RAG systems and AI agents.

Prompt Engineering

Prompt engineering is the practice of designing, testing, and iterating on the instructions given to a large language model to reliably produce accurate, consistent, and useful outputs. Well-engineered prompts can increase LLM task accuracy by 20-50% compared to naive instructions, often eliminating the need for more expensive fine-tuning.

Agentic AI

Agentic AI refers to AI systems that operate autonomously over extended task sequences -- planning actions, invoking tools, observing results, and re-planning until a goal is complete without step-by-step human guidance. Unlike single-turn chatbots, agentic systems can execute workflows that span minutes or hours, touching multiple APIs, databases, and services.

← Back to full glossary

Need help implementing this in your business?

Code and Trust translates AI concepts like hallucination into working implementations — starting with a workflow audit that shows exactly where it creates ROI.

Schedule AI Audit →