The paper introduces a taxonomy of security risks in cloud-hosted tool-enabled AI agents arising mainly from over-privileged tools and authority leakage, supported by scenarios, mitigations, and a small experiment.
The emerged security and privacy of llm agent: A survey with case studies
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
UNVERDICTED 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
Security Risks in Tool-Enabled AI Agents: A Systematic Analysis of Privileged Execution Environments
The paper introduces a taxonomy of security risks in cloud-hosted tool-enabled AI agents arising mainly from over-privileged tools and authority leakage, supported by scenarios, mitigations, and a small experiment.