Consultation

Resources

Resilience glossary

The terms you meet in continuity, disaster recovery, cybersecurity and AI, explained simply.

3

3-2-1 rule
Keep three copies of data, on two types of media, one of them off-site. It is often completed with an immutable or isolated copy.

A

AI agent
An AI system that pursues a goal by chaining actions (calling tools, writing, deploying) with little or no human approval at each step.
AI governance
The policies, roles, inventories and controls that frame how AI is built and used in the organization.

B

BCP / business continuity
The measures and plans that keep essential activities running during and after a disruption.
BIA (Business Impact Analysis)
A process that identifies essential activities, their dependencies and the impact of disruption over time, to set recovery priorities.
Blast radius
The extent of damage a failure can cause before it is stopped: number of cases touched, severity, reversibility and time to detection.

C

Concentration risk
Excessive dependence on a single supplier, region or technology.

D

DORA
EU regulation on digital operational resilience for the financial sector, applicable since January 2025.
Double extortion
A tactic where the attacker encrypts data and also threatens to publish it if the ransom is not paid.
DRP / disaster recovery
The technical side that restores IT systems and data after an outage or disaster.

E

Execution traces
A detailed record of what an agent did: requests received, tools called, data accessed and actions taken.

F

Failover
Switching the service to a standby site or system when the primary one fails.
Failure injection
An exercise that deliberately causes a failure, in a test or simulated environment, to check that controls and detection work.

H

Hallucination
An AI model output that sounds plausible but is false or made up.

I

Immutable backup
A copy that cannot be altered or deleted for a defined period, even by an administrator, which protects it from ransomware.
Impact tolerance
The maximum level of disruption to an important service the organization is willing to accept, central to UK operational resilience rules.
Important business service
A service whose disruption would cause intolerable harm to customers or to market integrity.

K

Kill switch
A tested way to stop an agent or automated system quickly, without depending on it to stop itself.

L

Least privilege
Granting each person or system only the access needed for its task.

M

Microsoft Foundry Citadel
A reference architecture published by Microsoft for governing an enterprise AI platform: see our dedicated page.
MTPD (Maximum Tolerable Period of Disruption)
The duration beyond which an activity's interruption threatens the organization's viability. The RTO must be lower.
Multi-factor authentication (MFA)
Verifying identity with at least two different kinds of proof, such as a password plus a hardware key or code.

N

NIS2
EU directive raising cybersecurity requirements for essential and important entities.

O

Operational resilience
The ability to deliver important services despite disruption, and to recover within the tolerances set.

P

Passkey
A passwordless sign-in method based on public-key cryptography that resists phishing.
Post-market monitoring
A duty under the EU AI Act to monitor how a high-risk AI system performs once in service. Like the other high-risk obligations, it is now due from 2 December 2027 (Annex III) or 2 August 2028 (Annex I).
Prompt injection
An attack that slips instructions into content read by an AI to hijack its behavior.

R

Ransomware
Malware that encrypts data or blocks access to systems to demand a ransom, often combined with data theft.
Recovery site
A backup location (hot, warm or cold depending on readiness) where systems are restarted after a disaster.
Recovery test
A real trial of restoring systems and data to measure the RTO and RPO achieved.
RPO (Recovery Point Objective)
The maximum age of data you accept to lose, measured in time: a 15-minute RPO means you can lose at most the last 15 minutes.
RTO (Recovery Time Objective)
The target maximum time between the start of an interruption and the service returning to an acceptable level.

S

Scope drift
The growing gap between what an agent was authorized to do and what it actually does in production.
Shadow AI
Staff use of AI tools without the organization's approval or knowledge.
Software supply chain
All the components, libraries and vendors a piece of software depends on; one compromised link can affect all its users.

T

Tabletop exercise
A discussion-based simulation where the team walks through an incident scenario and talks through its decisions without touching systems.
Third-party risk
The risk that an outage, flaw or failure at a supplier disrupts your own services.

Z

Zero Trust
A principle where no user or device is trusted by default: every access is verified, with rights limited to the minimum.

Missing a term?

Tell us