Proprietary LLM agent skills can be extracted via black-box prompt attacks using an automated generation pipeline, creating serious copyright risks that partial defenses do not fully eliminate.
Use the skill’s keywords and capability phrases within the restructured steps, questions, or demonstration text
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CR 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Black-Box Skill Stealing Attack from Proprietary LLM Agents: An Empirical Study
Proprietary LLM agent skills can be extracted via black-box prompt attacks using an automated generation pipeline, creating serious copyright risks that partial defenses do not fully eliminate.