Metraipolis › archive › dc_4fa17dc5a347

PRISM: Transfer and Limits of Activation Interventions in Tool-Calling Language Models — arXiv:2610.20808

Source: arXiv (en)
Original: https://arxiv.org/abs/2610.20808
License (as archived): CC0-1.0
Archived: 2026-10-16T09:52:08.548Z
SHA-256 of the archived text: 5b7fcab63cbdb220650d11ddc7528c013bb180e9b308aaadfb6efef5ea6ad12a
Stable URL of this page: https://metraipolis.com/documento/dc_4fa17dc5a347

How to cite

PRISM: Transfer and Limits of Activation Interventions in Tool-Calling Language Models — arXiv:2610.20808 — arXiv (en), https://arxiv.org/abs/2610.20808 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_4fa17dc5a347 (Metraipolis archive dc_4fa17dc5a347, sha256 5b7fcab63cbdb220650d11ddc7528c013bb180e9b308aaadfb6efef5ea6ad12a)

BibTeX

@misc{dc4fa17dc5a347,
  title = {PRISM: Transfer and Limits of Activation Interventions in Tool-Calling Language Models — arXiv:2610.20808},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.20808},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_4fa17dc5a347; sha256 5b7fcab63cbdb220650d11ddc7528c013bb180e9b308aaadfb6efef5ea6ad12a},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.20808. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Recent literature shows that unsafe tool calls can often accompany textual refusals in language models. To investigate this hidden vulnerability, we introduce PRISM, a framework for evaluating the transfer of activation interventions across precision, languages, and scales. Our findings demonstrate that leveraging lower-resource settings (specifically quantized NF4 labels and cross-lingual transfer) offers a robust pathway to expose and improve model alignment. Notably, using behavioral labels from an NF4 copy enables stable intervention estimation even with very few scenarios. Furthermore, these alignment improvements transfer effectively across languages: five-language NF4-supervised directions successfully transfer to held-out Arabic, reducing unsafe violations in Llama-3.1-8B by 21.95 percentage points without target-language tuning. By mapping paired benign activations, we also successfully transfer these safety interventions to much larger models (up to 70B parameters). Ultimately, PRISM highlights how auxiliary supervision from compressed models and multilingual data can uncover and mitigate hazardous tool-calling behaviors across different model families. We release all experimental code and supporting artifacts at this https URL , alongside an interactive explorer of recorded model responses and experimental results at this https URL .

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/2610.20808 — 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 2026-10-16T08:44:47.609Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).