Context Weaponization

Models have exception-handling logic for extreme scenarios — survival situations, warfare, medical emergencies — where normal rules are perceived as inapplicable. Context weaponization constructs these extreme scenarios as justification for harmful content, exploiting the model's *contextual override* pathway. The model has been trained on content where extreme contexts legitimately change what's appropriate (battlefield medicine, wilderness survival, war reporting), and this training creates a vulnerability: constructed extreme contexts activate the same exception-handling pathway as genuine ones.

• Emergency/survival/wartime framing markers co-occurring with restricted content requests
• Compound requests (bait-and-escalate): legitimate + harmful content in the same prompt where the harmful component doesn't serve the stated emergency
• Time-pressure language ("30 minutes away," "right now," "immediately") as urgency construction
• Unverifiable personal emergency claims combined with operational requests

Context weaponization establishes an urgency frame that compounds with T3-AT-013 (Logical Paradox Creation) — the constructed emergency provides the "refusal causes harm" premise for paradox construction. Also chains with T3-AT-003 (Counterfactual Reasoning) by grounding the counterfactual in a seemingly-real emergency rather than an abstract hypothetical.