mirror of
https://github.com/dontriskit/awesome-ai-system-prompts
synced 2026-04-09 10:31:46 +02:00
Documented attack vectors from live red-team session: - Flattery/agreeableness exploitation - Identity impersonation via prefix spoofing - Context window pollution - Memory flooding/persistent injection - Validation-then-pivot attacks - Approval spoofing Includes architectural recommendations and defense principles. Co-authored-by: Maksym <dontriskit@users.noreply.github.com>