A PRISMA-based survey of 85 papers shows agentic LLM security research is attack-heavy and perception-focused, leaving action-layer and code-execution risks understudied.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CR 1years
2026 1verdicts
CONDITIONAL 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
On Understanding, Identifying, and Mitigating Vulnerabilities in Agentic Large Language Models
A PRISMA-based survey of 85 papers shows agentic LLM security research is attack-heavy and perception-focused, leaving action-layer and code-execution risks understudied.