For Heads-of · Practitioner
Large Language Model (LLM)
A deep learning model trained on very large amounts of text, able to generate and interpret human-like language across a wide range of topics and tasks.
- LLM
- generative AI
- language models
- transformers
Definition
Large language models are transformer-based models with billions of parameters or more. They learn statistical patterns from their training data, which lets them complete text, answer questions, summarise, write code and hold a dialogue.
Scale
Larger models generally perform better on complex tasks, but cost more to train and run. Most organisations use them through a provider's API or inside a vendor's product, which makes them a third-party dependency.
Governance scope
LLMs bring a broad set of risks: they can produce harmful content, reproduce bias, state falsehoods fluently (misinformation and confabulation), and follow instructions hidden in the content they read (prompt injection). Their behaviour also changes between versions. Output validation and guardrails sit between the model and anything that acts on its output.
Related Terms
Agentic AI
An AI system that pursues a goal by deciding for itself which actions to take, in what order, using tools and other systems it has been given access to, with a person checking some or none of the steps. The governance change is that the unit of risk stops being a single output and becomes a sequence of actions with consequences.
Context Window
The maximum length of text (measured in tokens) that a language model can consider at once when generating a response.
Hallucination
Output that is fluent, confident and false. It arises from the same mechanism that produces correct output (a model predicting plausible continuations), which is why it is best understood as a property of the technology to be bounded rather than a defect to be fixed.
Prompt Injection
An attack in which untrusted content reaching a language model is interpreted as instructions rather than data, causing the model to act on an attacker's intent instead of the operator's. It is a property of how the application assembles its context, not a flaw in any particular model.
Token
A unit of text that a language model processes, typically a word, part of a word or a character. Models work on sequences of tokens, not raw text.
Transformer
A deep learning architecture that uses self-attention to process sequences of data, and the foundation of modern large language models.
Related risks and controls
Prompt injection
The attack class that exists because an LLM cannot reliably tell instructions from data.
Misinformation and confabulation
The failure mode that follows from generating plausible text rather than retrieving true text.
Output validation and guardrails
The control that treats model output as untrusted, which for an LLM is the only honest starting position.