The Complete AI Glossary for 2026

The AI vocabulary changes fast. This glossary covers the terms that came up in real 2026 meetings, articles and demos.

Table of contents

• Core terms
• Models
• Workflow
• Safety
• FAQ

Core terms

• LLM — Large language model
• Token — Chunk of text a model reads or writes
• Context window — How much a model can 'see' at once
• Prompt — Your instructions to the model
• Hallucination — Confidently wrong output

Model terms

• Foundation model — Large pre-trained model, base for others
• Fine-tune — Retrain on your data
• RLHF — Reinforcement learning from human feedback
• Distillation — Smaller, faster copy of a big model

Workflow terms

• Agent — Model plus tools plus loop
• RAG — Retrieval-augmented generation
• Tool use — Model calling external functions
• Eval — Scored test to measure quality

Safety terms

• Jailbreak — Prompt that bypasses safety
• Alignment — Making models act as intended
• Guardrail — Rule that constrains behavior
• Red team — Adversarial testing

FAQ

How often does the vocabulary change?

Every few months. We refresh this page quarterly.