Empirical attribution shows refusal blocks jailbreaks and prompt leakage, budget blocks sensitive disclosure and unbounded consumption, full stack needed for excessive agency, with refusal brittle to paraphrasing but budget robust.
Mitigating the OWASP top 10 for large language models applications using intelligent agents, 2026
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
-
Which Defense Closes Which Threat? Attributing OWASP-LLM-Top-10 Coverage and Its Brittleness Under Paraphrasing
Empirical attribution shows refusal blocks jailbreaks and prompt leakage, budget blocks sensitive disclosure and unbounded consumption, full stack needed for excessive agency, with refusal brittle to paraphrasing but budget robust.