Metraipolis › archive › dc_8543aeb2ed6c

HaloGuard 2.0: An Open-Weights Constitutional Output Guard for Multilingual Response Safety — arXiv:2611.02134

Source
arXiv (en)
Original
https://arxiv.org/abs/2611.02134
License (as archived)
CC0-1.0
Archived
2026-11-03T11:10:43.383Z
SHA-256 of the archived text
070fb8962e1781af560350e97ffe81c718dc083311a7243f6943a1255ad88dde
Stable URL of this page
https://metraipolis.com/documento/dc_8543aeb2ed6c

How to cite

HaloGuard 2.0: An Open-Weights Constitutional Output Guard for Multilingual Response Safety — arXiv:2611.02134 — arXiv (en), https://arxiv.org/abs/2611.02134 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_8543aeb2ed6c (Metraipolis archive dc_8543aeb2ed6c, sha256 070fb8962e1781af560350e97ffe81c718dc083311a7243f6943a1255ad88dde)
BibTeX
@misc{dc8543aeb2ed6c,
  title = {HaloGuard 2.0: An Open-Weights Constitutional Output Guard for Multilingual Response Safety — arXiv:2611.02134},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2611.02134},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_8543aeb2ed6c; sha256 070fb8962e1781af560350e97ffe81c718dc083311a7243f6943a1255ad88dde},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.02134. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

An input guard decides whether a request reaches the model, but never sees what the model says. Harm that emerges during generation, because a jailbreak passed the input layer or a benign request drifted into operational detail, is visible only on the output side and only as tokens arrive. We present HaloGuard 2.0, an open-weight streaming output guard that assigns a safe/unsafe probability to every response token in a single forward pass, so a deployment can stop the stream once harm is confirmed. The guard reuses the constitution of HaloGuard 1.0, our input guard, to organise a 478,532-response training corpus, and supervises each token from the point where the response becomes harmful. We find that a guard trained on conventionally shaped data learns a shortcut specific to streaming: it judges the prompt instead of the response, and handles only 13.9% of matched safe/unsafe response pairs correctly. We correct this with a second phase on 106,043 counterfactual responses built around the base guard's own failure modes, and release the equal-weight average of the base and corrective models. On six English response-safety benchmarks, HaloGuard 2.0-0.8B reaches a macro F1 of 83.6 (83.3 on held-out halves of the decision-rule search), up from 72.8, with 38% fewer false positives. It matches the strongest full-response guards with 5-10x fewer parameters and is 2.6 points above the strongest streaming guard, Qwen3Guard-Stream-4B. A multilingual variant, HaloGuard 2.0-Poly-0.8B, trained on PolyGuardMix, reaches 83.1 F1 on the response labels of PolyGuardPrompts across 17 languages, 5.0 points above the strongest baseline. On one H100 the guard costs 0.37 ms per token per stream when moderating 64 concurrent responses. Both guards are hardened continuously through the weekly red-teaming protocol of HaloGuard 1.0, extended here to the output layer.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.CL, nuovi lavori del giorno).
Categorie arXiv: Computation and Language (cs.CL) ; Cryptography and Security (cs.CR).
Pagina del lavoro: https://arxiv.org/abs/2611.02134 — 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 2026-11-03T10:59:52.326Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).