TRiSM-guided agentic workflows reduced RAG poisoning attack success from 31% to 10%, data-field injection from 42% to 25%, eliminated network injection, and raised report accuracy from 72.5% to 86.5% across five LLMs and 800 generations.
Language model inversion,
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
-
Why Trust Your Agent? Empirical Security Gains from TRiSM-Guided Agentic Workflows in Healthcare
TRiSM-guided agentic workflows reduced RAG poisoning attack success from 31% to 10%, data-field injection from 42% to 25%, eliminated network injection, and raised report accuracy from 72.5% to 86.5% across five LLMs and 800 generations.