paperAugust 2026Unreviewed

Decoding-Level Taboo: A Diagnostic Stress Test for LLM Robustness

T. Kamijo, Ori Rottenstreich, Javier Conde, Gonzalo Martínez, Pedro Reviriego

Abstract

Large language model evaluations typically focus on performance under nominal conditions, creating an illusion of capability where models comfortably walk a narrow, highly optimized generation corridor. In real-world deployments, however, complex system prompts, safety guardrails, and structural constraints continuously force models off this nominal path, driving a divergence between benchmark scores and deployment performance. To address this issue, we introduce Decoding-Level Taboo, a zero-pro

Cite

@misc{kamijo2026decodinglevel,
  title = {{Decoding-Level Taboo: A Diagnostic Stress Test for LLM Robustness}},
  author = {T. Kamijo and Ori Rottenstreich and Javier Conde and Gonzalo Martínez and Pedro Reviriego},
  year = {2026},
  month = aug,
  eprint = {2608.09900},
  archivePrefix = {arXiv},
  url = {https://www.semanticscholar.org/paper/64c109cad907daf65636234e2eed04e0fba80d70}
}