Yuanbo Zhou
1 indexed paper
Recent (6 mo)
1With code
0Influential cites
0Benchmarked
0Publications per year
126
Top categories
Crypto×1
Frequent co-authors
Research Timeline
2026
Prompt Overflow: What the Guardrail Inspects Is Not What the Model Infers
The paper introduces the Prompt Overflow Attack, demonstrating that guardrail models inspecting truncated or segmented inputs fail to detect malicious instructions that are only actionable when the full, overlong context is provided to the downstream LLM.
Highlighted terms show continued research focus across papers
Papers
cs.CRRecentMay 22, 2026
Prompt Overflow: What the Guardrail Inspects Is Not What the Model Infers
Yuanbo Zhou, Changjia Zhu, Junyu Wang, Xu He +4 more
The paper introduces the Prompt Overflow Attack, demonstrating that guardrail models inspecting truncated or segmented inputs fail to detect malicious instructions that are only actionable when the fu…
View →