Metraipolis › archive › dc_1237e48ed9f5

Alignment of LRMs via Counter-Aligned Few-Shot Conversation Exposure — arXiv:2609.27763

Source: arXiv (en)
Original: https://arxiv.org/abs/2609.27763
License (as archived): CC0-1.0
Archived: 2026-10-08T09:11:31.922Z
SHA-256 of the archived text: 51aca89539f1a3819879d7a505576dd629b78f72f285f82846aee4721e8b8883
Stable URL of this page: https://metraipolis.com/documento/dc_1237e48ed9f5

How to cite

Alignment of LRMs via Counter-Aligned Few-Shot Conversation Exposure — arXiv:2609.27763 — arXiv (en), https://arxiv.org/abs/2609.27763 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_1237e48ed9f5 (Metraipolis archive dc_1237e48ed9f5, sha256 51aca89539f1a3819879d7a505576dd629b78f72f285f82846aee4721e8b8883)

BibTeX

@misc{dc1237e48ed9f5,
  title = {Alignment of LRMs via Counter-Aligned Few-Shot Conversation Exposure — arXiv:2609.27763},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.27763},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_1237e48ed9f5; sha256 51aca89539f1a3819879d7a505576dd629b78f72f285f82846aee4721e8b8883},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.27763. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Large Reasoning Models (LRMs) rely on explicit chain-of-thought (CoT) reasoning and large context windows to achieve strong performance on complex tasks, but these features also introduce new attack surfaces. We show that LRMs' reasoning processes can be systematically steered by prepending counter-aligned few-shot conversations containing explicit CoT traces, leading to unsafe generations on harmful queries and unwarranted refusals on benign ones. We formalize this attack as SRCF (Steering Reasoning via Counter-Aligned Few-shot Conversations) that operates solely through a flexible conversational interface and requires no access to the model's parameters and gradients. Our key insight is that SRCF exploits an adversarial generalization issue that induces a representation drift, causing the representations of benign and harmful inputs to shift in a similar direction. This observation motivates our post-training defense, ARCF (Aligning Reasoning via Counter-Aligned Few-Shot Conversations), which exposes models to counter-aligned conversational contexts while enforcing aligned targets. ARCF is compatible with existing post-training methods and consistently improves safety and helpfulness without degrading utility.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-24 in cs.AI).
Categorie arXiv: Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.27763 — 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 2026-10-08T08:13:26.534Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-24?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).