Skip to content

Stockertim Dictionary · Artificial intelligence · Cybersecurity

Defence in depth

Also called layered defence, layered security

In short

Defence in depth is the practice of using several independent layers of protection, so that one failure does not expose everything.

  • Intermediate
  • Concept
  • Checked

In detail

Defence in depth means not relying on any single control. Instead, an organisation stacks several protections, each able to stop or catch a problem the others miss. A building might have a locked gate, a staffed reception, door passes and cameras; a computer system has similar layers, from strong log-ins to permissions, monitoring and backups.

For AI tools the layers often include input handling that marks outside content as untrusted, a model configured to refuse obvious misuse, limited permissions for any tools it can call, human approval for high-impact actions, checks on outputs before they reach other systems, logging and alerts, and a tested plan for when something goes wrong.

The approach matters because every control fails sometimes. Filters miss cleverly worded content, people approve things in a hurry, and settings drift. When layers are truly independent, the chance that all of them fail at the same moment is far smaller than the chance that any one fails. If layers share a weakness, such as one admin account that can switch them all off, the benefit shrinks.

At work, defence in depth explains why staff are asked to follow several rules at once, such as using approved tools, checking outputs and confirming payments by phone.

Worked example

An illustrative calculation of three independent layers, each of which misses 10% of harmful requests.

  1. Chance one layer misses: 10% = 0.1
  2. Chance all three miss, if independent: 0.1 x 0.1 x 0.1 = 0.001, which is 0.1%
  3. If the layers share a weakness and fail together, the chance stays near 10%

Independent layers reduce the chance of a full miss from 10% to 0.1% in this illustration, which is why independence matters.

In practice

Security teams design AI deployments with several separate controls, test each one, and avoid single points of failure such as one shared key or one admin who can disable everything.

Watch out

  • More layers are not always better if they all depend on the same thing; independence is the point.
  • Layered defences reduce risk but never remove it, so monitoring and incident response stay necessary.

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