LLMs for code vulnerability detection show average susceptibility of 33.2% to framing, 23.5% to anchoring, and 18.4% to halo effects, with a black-box attack suppressing up to 97% of detections.
When wording steers the evaluation: Framing bias in llm judges,
3 Pith papers cite this work. Polarity classification is still indexing.
3
Pith papers citing it
years
2026 3verdicts
UNVERDICTED 3representative citing papers
Introduces Fragile benchmark showing LLMs have 28.6% average decision flip rate under fact-preserving framing and proposes Valign method to reduce flips via value-anchored hidden-state steering and projection.
Position paper calling for stronger evidentiary standards and a diagnostic checklist in anthropomorphic misalignment research.
citing papers explorer
-
Words Speak Louder Than Code: Investigating Cognitive Heuristics in LLM-Based Code Vulnerability Detection
LLMs for code vulnerability detection show average susceptibility of 33.2% to framing, 23.5% to anchoring, and 18.4% to halo effects, with a black-box attack suppressing up to 97% of detections.
-
Framing Matters: Addressing Framing Sensitivity in Decision-Making through Behaviorally-Grounded Value Alignment
Introduces Fragile benchmark showing LLMs have 28.6% average decision flip rate under fact-preserving framing and proposes Valign method to reduce flips via value-anchored hidden-state steering and projection.
-
Position: Anthropomorphic Misalignment Research Needs Stronger Evidence
Position paper calling for stronger evidentiary standards and a diagnostic checklist in anthropomorphic misalignment research.