Canon term · Context & memory

Prompt Injection

An attack exploiting how an application combines untrusted input with a prompt written by a higher-trust party, causing the system to follow the untrusted instructions.

  • StatusIndustry term
  • AdoptionWidely used
  • LayerGovernance

Why it matters

In an agentic setting this is not a content problem, it is an authority problem: a successful injection causes an agent to exercise real permissions on an attacker's behalf. The defense is the authority envelope, not better filtering.

Where it comes from

Defined in the AIEOG Shared AI Lexicon (February 2026), adapted from NIST.

Defining sources

How it connects

What it needs

MARQUE questions it addresses

  • AuthorityWhat is the agent permitted to commit, and who granted that?

Connected through the MARQUE (a reference framework organized around eight questions) crosswalk, an editorial mapping. These objectives from the Financial Services AI Risk Management Framework are assigned to the same questions that this term addresses. The ones sharing most of this term's vocabulary are shown first, as a reading aid. The assignment is to the question, not to the term: the source does not map this term to any objective directly.

Question: Authority · 39 control objectives assigned

  • GV-2.2.2Mandatory AI Risk Management TrainingGovern
  • MP-3.3.1AI Application Scope DefinitionMap
All 39 objectives assigned to Authority

Cite and reuse

Cite the identifier QIS-TERM-00012, not the URL: identifiers are stable, URLs may change. This entry is free to read, quote and index under the dual license. Corrections sent to editor@qisfund.com are published.

Technical details: identifier, graph relationships and record
QIS-TERM-00012: outbound relationships as published in api/graph.json and api/lexicon.json
PredicateTarget identifierTargetPredicate meaning
addressesmarque:authorityAuthorityThe subject speaks to the object as a question or concern.
constrainedByterm:authority-grantAuthority GrantRead the other way round: the object sets limits on the subject. The subject bounds what the object may do.
Canonical identifier
QIS-TERM-00012
Tier
Canonical industry term
Domain · layer
Context & memory · Governance
Semantic aliases
None recorded
First published
2026-08-02
Last reviewed
2026-08-02 · 180-day cycle