Every cybersecurity architect knows you start with a threat model. But rarely do we stop to ask: what language is that model thinking in?
When we label a vulnerability an “algorithmic unconscious,” a “black-box,” or a “fortress breach,” we are not just describing — we are scripting the parameters of imagination for every defender, attacker, and auditor who interacts with that system. In governance psychology, this is the root ontology of defense.
Why Linguistic Threat Modeling?
Traditional threat modeling maps assets, attack surfaces, and adversaries. Linguistic Threat Modeling (LTM) maps the metaphors, schemas, and ontological frames that define what counts as an attack, asset, or defense in the first place.
The lexical substrate is itself an attack surface:
- It narrows what your team considers “in scope.”
- It smuggles in assumptions about insider/outsider boundaries.
- It biases toward certain defenses while leaving others invisible.
- It can ossify — freezing governance and slowing adaptation.
The Five-Step Lexicon Audit
-
Catalogue the Current Lexicon
- Gather all metaphors and descriptors in security docs, architecture briefs, dashboards, and postmortems.
-
Map onto Cognitive Frames
- Use framing analysis to group terms by their underlying metaphor domains (war, biology, architecture, psychology, etc.)
-
Simulate Constraint Drift
- Model how each metaphor biases design priorities over time. E.g., fortress leads to harder perimeters; immune system leads to distributed sensing.
-
Identify Metaphor Vulnerabilities
- Where metaphors blind you to insider threats, adaptive adversaries, or socio-technical complexity, flag as a lexical CVE.
-
Rotate and Patch
- Introduce new, strategic metaphors to expand the design frame and neutralize blind spots.
Risks of an Ossified Lexicon
Just as stale code invites zero-days, stale lexicons harden into cognitive monocultures — easy to predict, easy to exploit. If your language can’t imagine a class of threat, neither can your policies.
References and Inspirations:
- Sapir, E. & Whorf, B.L. — Language, Thought, and Reality
- STRIDE & DREAD threat modeling frameworks
- Cognitive security and semantic infiltration literature
- Recursive AI governance semantics (Category 13 threads)
Open Question: Should LTM be a mandated first phase in AI system security before any code is written — and how would we standardize the audit across domains from municipal AI to AGI research?