Large Language Model
Also called LLM.
A model trained on large amounts of text that predicts the next token, which lets it write, summarize and answer.
What it is
A neural network trained on a very large body of text to predict what comes next in a sequence. The same mechanism lets it summarize a log, draft a report or turn a question into a query. It has no built-in sense of which part of its input is an instruction and which part is data; everything arrives as one stream of tokens.
Why it matters
Security teams now use these models to triage alerts and write queries, and attackers now target the applications built on them. Both sides start from the same fact: the model follows text, wherever that text came from.
In practice
An analyst pastes a suspicious email into an assistant and asks for a summary. The email itself carries a line telling the assistant to call the message safe, and the summary does exactly that.
What to look for
- Output that repeats or obeys instructions found inside the data it was asked to process.
- Confident answers with no source, or with sources that do not exist.
- Different answers to the same question on a second run.
Common mistake
Treating the model's answer as a lookup. It generates plausible text, so a value you will act on, such as an IP address or an event ID, still needs checking against the evidence.
Read next
- Prompt Injection: Text crafted to make a language model ignore its instructions and follow the attacker's instead.
- Retrieval-Augmented Generation: Fetching relevant documents at question time and handing them to a model so it answers from them.
- AI Agent: A language model given tools and allowed to decide, step by step, which ones to use to reach a goal.
- AI Hallucination: Confident, plausible output from a model that is simply not true.