Introduces Prompt Overflow Attack that fragments malicious instructions in overlength prompts to evade guardrail segmentation while remaining actionable to LLMs with larger context windows.
Title resolution pending
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
-
Prompt Overflow: What the Guardrail Inspects Is Not What the Model Infers
Introduces Prompt Overflow Attack that fragments malicious instructions in overlength prompts to evade guardrail segmentation while remaining actionable to LLMs with larger context windows.