Metraipolis › archive › dc_e64685aa36b7

RealControlBench: A Benchmark for Interpretability-Informed Fine-Grained Control of LLMs and Agents — arXiv:2610.27295

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.27295
License (as archived)
CC0-1.0
Archived
2026-10-21T07:08:52.903Z
SHA-256 of the archived text
27cb1cf298916e913b032ecb5dd6a011123c07c60334bd87e775e829e183ca17
Stable URL of this page
https://metraipolis.com/documento/dc_e64685aa36b7

How to cite

RealControlBench: A Benchmark for Interpretability-Informed Fine-Grained Control of LLMs and Agents — arXiv:2610.27295 — arXiv (en), https://arxiv.org/abs/2610.27295 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_e64685aa36b7 (Metraipolis archive dc_e64685aa36b7, sha256 27cb1cf298916e913b032ecb5dd6a011123c07c60334bd87e775e829e183ca17)
BibTeX
@misc{dce64685aa36b7,
  title = {RealControlBench: A Benchmark for Interpretability-Informed Fine-Grained Control of LLMs and Agents — arXiv:2610.27295},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.27295},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_e64685aa36b7; sha256 27cb1cf298916e913b032ecb5dd6a011123c07c60334bd87e775e829e183ca17},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.27295. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Existing steering benchmarks are built around simplified, isolated generation tasks and evaluate a single behavioral attribute under controlled settings that do not reflect the diversity and complexity of practical deployment. As a result, performance on these benchmarks provides only a limited picture of how steering methods perform in the real world. We therefore introduce RealControlBench, a comprehensive benchmark for evaluating fine-grained steering of LLMs and LLM agents across diverse behavioral targets and interaction settings. RealControlBench extends steering evaluation beyond isolated generations to complex, execution-grounded, multi-turn agent interactions, covering five complementary dimensions: instruction-following control, agentic tool-use decision control, reasoning-performance steering, reasoning-length control, and trajectory-conditioned safety awareness. Rather than measuring steering success alone, the benchmark jointly evaluates target controllability and capability preservation, including instruction following, reasoning performance, and downstream agent utility and safety. It further examines whether steering effects remain specific and robust across prompt variations, domains, and model architectures. Our evaluation shows that existing steering methods remain unreliable across realistic control objectives: they often fail to consistently induce the intended behavior and can substantially degrade the model's underlying task performance. These results suggest that strong steering effects in controlled settings do not necessarily translate into reliable control in more realistic evaluations. By providing a unified, model-agnostic steering evaluation framework, RealControlBench aims to support the development of steering methods that are not only effective but also precise, robust, and useful in realistic AI deployments.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.CL, nuovi lavori del giorno).
Categorie arXiv: Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2610.27295 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-21T07:05:58.109Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).