SOURCE-LINKED INTELLIGENCE

Overflip: Repetition-Induced Label Flips in Guardrail Models

arXiv · AI, language, vision and robotics · article · Sep 14, 2026 · UTC

Guardrail models are classifiers deployed to screen malicious prompts and responses in LLM-based services. To meet latency constraints, many lightweight guardrails adopt compact Transformer backbones (e.g., DeBERTa) that are trained with short context windows (typically 512 tokens) and rely on bucketed relative positional encodings to process longer inputs. Prior evaluations assume that a guardrail's decision is stable as the input is lengthened. We show that this assumption can fail. We identify Overflip, a repetition-induced instability where repeating a prompt causes the guardrail's predict

Read original source ↗ Open in workspace

recordType

paper

region

Global