Metraipolis › archive › dc_4a67e0caaa9c
SteerCheck: Attribution Specificity and Alignment Leakage in Activation-Steering Audits — arXiv:2608.24335
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2608.24335
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-18T06:18:05.591Z
- SHA-256 of the archived text
f5ae14eb9795f424546df15b0d8736e74cce2ff3f5624ec889faba26d9b825ce- Stable URL of this page
https://metraipolis.com/documento/dc_4a67e0caaa9c
How to cite
SteerCheck: Attribution Specificity and Alignment Leakage in Activation-Steering Audits — arXiv:2608.24335 — arXiv (en), https://arxiv.org/abs/2608.24335 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_4a67e0caaa9c (Metraipolis archive dc_4a67e0caaa9c, sha256 f5ae14eb9795f424546df15b0d8736e74cce2ff3f5624ec889faba26d9b825ce)
BibTeX
@misc{dc4a67e0caaa9c,
title = {SteerCheck: Attribution Specificity and Alignment Leakage in Activation-Steering Audits — arXiv:2608.24335},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2608.24335},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_4a67e0caaa9c; sha256 f5ae14eb9795f424546df15b0d8736e74cce2ff3f5624ec889faba26d9b825ce},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.24335. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Activation steering can change behaviour without establishing that the effect is specific to the intended concept. We introduce SteerCheck, a preregistered attribution audit that matches off-target KL and separates mean, protected-tail, polarity, transfer, and semantic claims. Exact replay of 960 Qwen3-14B interventions reveals complementary limits of common controls: isotropic directions occupy a narrow near-orthogonal region, whereas sign-randomized same-construction directions often retain substantial target alignment. Effect is strongly associated with signed cosine within the sign-randomized family ($\rho=.94$); $25.3\%$ of its draws exceed cosine $.5$, and every draw exceeding the observed mean effect has cosine above $.80$. This alignment leakage does not by itself invalidate a conditional randomization test; it limits what the comparator can distinguish and motivates reporting exchangeability assumptions, a construction diagnostic $A$, and the empirical cosine distribution. The primary Qwen complete gate remains negative because the protected tail fails all families. On independent data, continuous margin transfers only in Qwen and accuracy transfers in no selected cell. Prospectively registered language controls pass the complete gate in Qwen and DeepSeek, while a passing DeepSeek detox comparator rules out categorical separation; all nominal passes are sensitive to $\Gamma=1.10$. Frozen three-rater open-generation evaluation supports factual correction in DeepSeek but not Qwen; the automatic judge fails calibration (macro-F1 $.562$), so null-wide semantic results remain descriptive. SteerCheck makes these conditional and mixed conclusions auditable. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-26 in cs.CL). Categorie arXiv: Computation and Language (cs.CL). Pagina del lavoro: https://arxiv.org/abs/2608.24335 — 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 2026-10-18T06:17:29.365Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-08-26?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).