Metraipolis › archive › dc_bd2244c768eb
Multi-Reward Hacking: From Aggregation-Induced Drift to Trust-Anchored Defense — arXiv:2610.21404
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2610.21404
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-16T09:19:40.322Z
- SHA-256 of the archived text
15a351e1d6f2d21bf4bd43f9052e9dfc204d8e0e1b8e19ea26c5436ddffc1a82- Stable URL of this page
https://metraipolis.com/documento/dc_bd2244c768eb
How to cite
Multi-Reward Hacking: From Aggregation-Induced Drift to Trust-Anchored Defense — arXiv:2610.21404 — arXiv (en), https://arxiv.org/abs/2610.21404 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_bd2244c768eb (Metraipolis archive dc_bd2244c768eb, sha256 15a351e1d6f2d21bf4bd43f9052e9dfc204d8e0e1b8e19ea26c5436ddffc1a82)
BibTeX
@misc{dcbd2244c768eb,
title = {Multi-Reward Hacking: From Aggregation-Induced Drift to Trust-Anchored Defense — arXiv:2610.21404},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2610.21404},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_bd2244c768eb; sha256 15a351e1d6f2d21bf4bd43f9052e9dfc204d8e0e1b8e19ea26c5436ddffc1a82},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.21404. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
To improve task performance while encouraging desirable response properties, LLM post-training often combines a primary reward (e.g., correctness) with auxiliary rewards (e.g., brevity or efficiency). However, how these rewards are aggregated can critically affect training outcomes. We show that symmetric aggregation, which treats reward channels equally, can induce a failure where aggregate reward improves while primary-task performance degrades, even when individual rewards are meaningful. We investigate this aggregation-induced failure through a one-step multi-reward decision formulation, characterizing when it constitutes multi-reward hacking. We prove that, under symmetric aggregation, deterministic conditional-mean GRPO converges to auxiliary-favoring behavior when auxiliary rewards are more reliably obtained, regardless of initialization. We propose Trust-Anchored Gating (TAG), which first elects an anchor channel based on the policy drift induced by optimizing each channel alone, then gates auxiliary credit on anchor satisfaction. Unlike calibration-sensitive static reweighting, TAG structurally eliminates credit for auxiliary-only outcomes while preserving additional credit for joint success. In the same analytical setting, TAG provably reverses this selection, converging to the primary-favoring action from any initialization when the primary channel is anchored. Experiments across Qwen and Llama models on mathematics and program synthesis reveal substantial degradation under symmetric, constrained, and normalized aggregation, whereas TAG mitigates this failure. On Qwen3-4B with a brevity reward, TAG achieves 0.581 pass@4 versus 0.220 for additive aggregation, while shortening responses by 44% relative to correctness-only training. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno). Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2610.21404 — 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 2026-10-16T08:44:17.946Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).