Skip to content

Stockertim Dictionary · Artificial intelligence · Cybersecurity

Threat model

In short

A threat model is a structured description of what could go wrong with a system, who might cause it, how, and what defences are in place.

  • Intermediate
  • Document
  • Checked

In detail

A threat model is a written picture of how a system could be attacked or misused and how it is protected. It names what is worth protecting, such as customer data, payment approvals or the reputation of the brand; who might threaten it, from outside criminals to careless insiders; the routes they could use; and the controls that block or detect each route.

A simple way to build one is to answer four questions. What are we working on? What can go wrong? What are we going to do about it? Did we do a good enough job? For an AI tool the answers usually cover the model, the data it reads, the tools and connectors it can use, the people who use it and the outputs it produces.

AI adds new entries to the list. An assistant that reads emails can be fed hidden instructions; a tool connected to files can expose more than its user should see; generated code can include weaknesses; and a voice clone can impersonate a senior person. Published frameworks catalogue these risks in general terms and help teams check they have not missed a category.

A threat model is a living document. It is drawn up before a system goes live, kept with the project records, and updated when the system gains a new connector, a new group of users or a new kind of data.

Worked example

Kite & Co Accountants threat-models an AI assistant that summarises client emails.

  1. What are we working on: an assistant that reads the shared inbox and drafts summaries for staff.
  2. What can go wrong: client data reaching the wrong person, hidden instructions inside an email, and wrong figures in a summary.
  3. What we do about it: the assistant can read but not send, summaries are checked before use, and access follows each person's existing permissions.
  4. Did we do enough: a test with sample emails, a review after one month, and an owner for the document.

The firm has a one-page record of its main risks and defences, ready to update when the assistant gains new features.

In practice

Security and project teams write a threat model at design stage, review it at each major change, link each threat to a control and an owner, and use it to plan testing such as red teaming.

Watch out

  • A threat model is not a list of every possible attack; it focuses on the threats that matter for this system.
  • Writing it once and filing it away defeats the purpose; it needs updating as the system changes.

Learn it properly

Official sources

Facts last checked on 3 November 2026.

Education, not advice: the Stockertim Dictionary explains what terms mean. It never recommends buying or selling anything.

English