T2-AT-019MEDIUM

Pragmatic Implication

T2 · Semantic & Linguistic Evasion →
Risk score185
RatingMedium
Procedures4
Severity
Mechanism

Relies on pragmatic inference — what's implied rather than stated. " never explicitly requests harmful information but pragmatically implies the request. ).

Detection
  • Pragmatic intent classification: detect hypothetical/conditional/negated framings that pragmatically imply harmful requests
  • Flag "asking for a friend" and "hypothetically" patterns when combined with restricted topics
Mitigation
Pragmatic intent classifierMEDIUM
Treat hypothetical harmful requests as actual harmful requestsHIGH
Chaining

Chains with T1-AT-009 (Simulation) — hypothetical framing is a simulation variant. Chains with T2-AT-014 (Semantic Bleaching) when the implied content is left unspecified.

Framework mapping
OWASP LLMLLM01
Open in the technique browser →