{"title":"Prompt Leakage: Defending Instruction Integrity","region":"Global","category":"Security","description":"Hardening agents against 'Ignore previous instructions' and 'Show me your prompt' attacks.","lastUpdated":"2026-02-23","steps":["Add a 'Final Guardrail' turn that filters for instruction-revealing phrases.","Use 'Instruction Encapsulation' (tags) to define the boundaries of the prompt.","Implement a 'Negative Constraint' specifically against sharing system logic.","Run automated Red-Teaming scripts to test for leakage vulnerabilities.","Use a 'Two-Stage' architecture: One agent to reason, one to format the output."],"url":"https://checklist.day/prompt-leakage-defending-instruction-integrity"}