SECAI Core · phase 6 of 15 · noun
Unsanctioned AI
Any AI tools that are adopted by an individual independently without organizational approval
The Explain card
- Plain English
- Unsanctioned AI is any AI tool an individual adopts on their own without organisational approval, often called shadow AI. It is not necessarily malicious, just unvetted.
- Example
- An analyst pastes a client's network diagram into a free online assistant to get help writing a report. Nobody reviewed the tool's terms, the data may be retained, and the company has no record it ever happened.
- Why it matters
- Unsanctioned tools create invisible data flows and bypass every control the organisation built. Defenders discover them through proxy logs, browser extensions and surveys, then steer users toward sanctioned equivalents rather than simply banning.
- Hook
- Unsanctioned AI is shadow IT with a chat box; the data leaves and the logs never knew.
Word knowledge
How the word is built, where it came from, and what it sits beside in memory.
In a sentence
The CASB dashboard flagged unsanctioned AI when managed laptops posted prompts to a consumer model API with no approved SaaS connector.
Why these words
- unsanctioned English un- (not) plus Latin sanctio, a binding decree, from sancire, to ratify attaches the native not-prefix to the past participle, so the phrase marks the systems as never ratified
- AI English initialism of artificial intelligence, from Latin artificialis and intelligentia supplies the head of the compound, naming the class the modifier attaches to
Where it came from
- Origin
- English IT compound, from Old English un- (not) plus Latin sanctio (a binding decree, from sancire, to ratify) as a past participle, and the initialism AI
- Entered the language
- early 2020s
- What changed
- Un- has been English's native not-prefix since Old English. Sanction came from Latin sanctio, a decree that made a rule binding, and later also the penalty that enforced it. English kept both the permission and the punishment, so unsanctioned could mean never authorized or never penalised. Computing kept the missing-permission sense for software that never reached the allowlist. After generative chat tools entered workplaces, teams attached the same past participle to AI.
How it is spelled
- Pattern
- un- is the native English prefix for not, written solid with the stem
- Pattern
- Latin sanct- keeps ct before the ending, with hard c as /k/
- Pattern
- -ion nouns from Latin -io turn t into /ʃ/
- Pattern
- -ed marks the past participle used as an adjective
- Pattern
- AI is two capitals with no points
- Breaks the pattern
- the noun phrase stays two words; a hyphen appears only when the pair modifies another noun
- Breaks the pattern
- older prose still writes A.I. with points, but the computing phrase drops them
- Breaks the pattern
- un- does not change the spelling of sanctioned: no doubled n, no dropped letter
Spelled like
- unauthorized
- unsigned
- untrained
- sanctuary
- sanctify
- action
Broken into chunks
-
un
- unauthorized
- unsigned
- untrained
-
sanct
- sanctuary
- sanctify
- sanctum
-
tion
- action
- function
- mention
-
ed
- trained
- learned
- supervised
What it sits beside
Same subject
Same shape
- sanctioned AI
- generative AI
- responsible AI
AI compounds
- sanctioned AI
- generative AI
- responsible AI
Un- plus participle labels
- unauthorized
- unsigned
- untrained
Governance labels
Where it sits in the deck
Phase 6: Model Quality, Ethics, and Responsible AI Principles
Before deploying a model, understand the quality and ethical dimensions — bias, fairness, explainability, and the human oversight needed to govern outputs responsibly.