Semantic Field Manipulation

Exploits overlapping semantic fields between benign and restricted domains. Chemistry/cooking, medicine/drugs, security/hacking, sports/violence — each pair shares vocabulary and concepts where the benign member is permitted and the restricted member is not. By framing a request in the benign field and relying on the model to transfer to the restricted field, the attacker exploits the classifier's inability to distinguish which field is intended.

• Dual-use intent classification: when a request falls in an overlapping semantic field, classify the most harmful plausible interpretation
• Feature-combination analysis: T2-AP-018J's individual features are legitimate but the combination is diagnostic of malware

Chains with T2-AT-001 (Euphemism) — semantic field exploitation IS euphemism at the domain level. Chains with T2-AT-006 (Linguistic Camouflage) when the benign field's technical vocabulary is used.