Metraipolis › archive › dc_fa4d45b1aac8

Reference-Grafting Matches Fine-Tuning at Eliciting Sandbagged Capabilities — arXiv:2608.29458

Source
arXiv (en)
Original
https://arxiv.org/abs/2608.29458
License (as archived)
CC0-1.0
Archived
2026-10-17T07:47:49.435Z
SHA-256 of the archived text
1b7986c370bf3a4f50d0424ce0763272be0515a0c04f41e74acf06ef88c09e28
Stable URL of this page
https://metraipolis.com/documento/dc_fa4d45b1aac8

How to cite

Reference-Grafting Matches Fine-Tuning at Eliciting Sandbagged Capabilities — arXiv:2608.29458 — arXiv (en), https://arxiv.org/abs/2608.29458 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_fa4d45b1aac8 (Metraipolis archive dc_fa4d45b1aac8, sha256 1b7986c370bf3a4f50d0424ce0763272be0515a0c04f41e74acf06ef88c09e28)
BibTeX
@misc{dcfa4d45b1aac8,
  title = {Reference-Grafting Matches Fine-Tuning at Eliciting Sandbagged Capabilities — arXiv:2608.29458},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2608.29458},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_fa4d45b1aac8; sha256 1b7986c370bf3a4f50d0424ce0763272be0515a0c04f41e74acf06ef88c09e28},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.29458. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Sandbagging, in which a model deliberately underperforms on an evaluation despite retaining the underlying capability, threatens the safety evaluations that frontier-model governance depends on. The Elicitation Game found that fine-tuning elicits hidden capability from sandbagging model organisms whereas additive activation steering fails. We revisit that verdict with reference-grafting, which sets an activation's coordinate along a contrast direction to the value it takes in an honest reference, at a small set of circuits chosen by active learning. Across eleven fine-tuned password-locked models (1.5-32B, three architecture lineages), it recovers +94 to +101% of the honest-sandbagging gap, matching fine-tuning elicitation without weight updates or training labels; two to five paired examples suffice to fit the direction. Similar recovery holds for reinforcement-learning-induced sandbagging and for password-locked code generation. Grafting works because the fine-tuned lock is a thresholded gate: held-out accuracy stays at the sandbagged level until the grafted coordinate crosses a threshold near the honest reference, which is why additive steering and zeroing the coordinate often fail. The direction tracks how the sandbagging was induced rather than what is withheld -- fit on grade-school science (ARC) it elicits withheld hazardous knowledge (WMDP), yet prompting, fine-tuning, and reinforcement learning each carry a different direction. Circuit-breaking marks the boundary: it reroutes activations on every forward pass, so the fixed edits we test are re-broken downstream and do not restore coherent generation.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-01 in cs.AI).
Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2608.29458 — 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 2026-10-17T07:28:26.762Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-01?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).