Metraipolis › archive › dc_5defe13fec5d

Semantic Fibers and Cross-Gram Interference: A Calculus of Safety Drift in Overcomplete Representations — arXiv:2609.14861

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.14861
License (as archived)
CC0-1.0
Archived
2026-10-15T22:53:25.172Z
SHA-256 of the archived text
3519e75206828423c89bbbe5ef1327b6f89d5ac08be1610dd98a9067055dd4f8
Stable URL of this page
https://metraipolis.com/documento/dc_5defe13fec5d

How to cite

Semantic Fibers and Cross-Gram Interference: A Calculus of Safety Drift in Overcomplete Representations — arXiv:2609.14861 — arXiv (en), https://arxiv.org/abs/2609.14861 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_5defe13fec5d (Metraipolis archive dc_5defe13fec5d, sha256 3519e75206828423c89bbbe5ef1327b6f89d5ac08be1610dd98a9067055dd4f8)
BibTeX
@misc{dc5defe13fec5d,
  title = {Semantic Fibers and Cross-Gram Interference: A Calculus of Safety Drift in Overcomplete Representations — arXiv:2609.14861},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.14861},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_5defe13fec5d; sha256 3519e75206828423c89bbbe5ef1327b6f89d5ac08be1610dd98a9067055dd4f8},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.14861. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

A deployed language model may refuse a harmful request in English yet comply with its faithful translation, revealing a cross-lingual safety failure that cannot be characterized reliably by output behavior alone. We formalize this phenomenon through an audited equivalence relation and show that, for a declared quotient, representation, metric, feature dictionary, scoring head, threshold, and contrast model, the resulting safety drift admits an exact linear-algebraic characterization. Specifically, the drift is a cross-Gram functional of the within-fiber contrast; its worst admissible value is a support function, while margin invariance is characterized by an annihilator condition. We introduce an intrinsic calibrated exposure measure, governed by the leverage duality $\chi^2=1/\ell-1$, which separates observed drift into three diagnostically distinct regimes: a reader fault removable by recalibration, an exact correction that is too ill-conditioned to be reliable, and a representation-level collision that no readout-only intervention can remove. Thus, identical observed exposure can lead to fundamentally different remediation verdicts. The framework also extends to cone-valued safety heads. An untied order-swap identity provides a diagnostic for the linear control interface; its calibration-state residual predicts a distinct three-control composition error on unseen states and targets, achieving median Spearman correlation $0.964$, compared with $0.269$ for a static cross-Gram baseline. etc.....

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-15 in cs.AI).
Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.14861 — 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 2026-10-15T22:29:03.747Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-15?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).