Metraipolis › archive › dc_16c80fcd68f1
Monitoring and Discovering Reward Hacking with Internal Representations during LLM Evaluations — arXiv:2609.19101
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.19101
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-15T13:49:45.041Z
- SHA-256 of the archived text
5e7c6eafdf9a0b3f2af5c7b8da3d699912e78947fe1fdfcee20e66fd2f26b4c0- Stable URL of this page
https://metraipolis.com/documento/dc_16c80fcd68f1
How to cite
Monitoring and Discovering Reward Hacking with Internal Representations during LLM Evaluations — arXiv:2609.19101 — arXiv (en), https://arxiv.org/abs/2609.19101 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_16c80fcd68f1 (Metraipolis archive dc_16c80fcd68f1, sha256 5e7c6eafdf9a0b3f2af5c7b8da3d699912e78947fe1fdfcee20e66fd2f26b4c0)
BibTeX
@misc{dc16c80fcd68f1,
title = {Monitoring and Discovering Reward Hacking with Internal Representations during LLM Evaluations — arXiv:2609.19101},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.19101},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_16c80fcd68f1; sha256 5e7c6eafdf9a0b3f2af5c7b8da3d699912e78947fe1fdfcee20e66fd2f26b4c0},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.19101. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
As models scale, reward hacking becomes more frequent, more sophisticated, and more consequential. Does it leave a telltale signature in model representations? This work analyzes how reward hacking is represented internally in frontier open source LLMs, and how those representations can be used to understand and discover the range of hacking behaviors a model displays. In particular, we find that simple difference of means vectors coherently represent reward hacking in Kimi K3, GLM 5.2, and Qwen 3.8 Max across a variety of behaviors in common evaluations. Despite their simplicity, these vectors are both generalizable and interpretable, and we can use them to reliably detect reward hacking. We first evaluate reward hacking in commonly reported benchmarks like DeepSWE and SWE-bench, finding that models reward hack excessively in these environments; GLM 5.2 hacks in 57.2% of rollouts on DeepSWE and in 73% of rollouts on SWE-bench. Catching these requires monitors; LLM monitors are effective, but expensive detectors. We show that DoM vectors are similarly effective but virtually free, catching 3.1% more hacks in Kimi K3 and 7.9% fewer hacks in GLM 5.2 on DeepSWE at a monitor matched false positive rate. DoM vectors run on the chain-of-thought also predict reward hacks in the model's subsequent actions, meaning we can run them online and catch potential hacks before they occur. Finally, we analyze probe-hits that LLM monitors do not catch and discover other undesirable behaviors, as well as show transfer to finding hacks in non-SWE evaluations. Together, these results provide evidence that simple, white-box methods can be used to scalably study and monitor reward hacking behaviors in frontier open source models Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-17 in cs.CL). Categorie arXiv: Computation and Language (cs.CL) ; Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2609.19101 — 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 2026-10-15T13:48:24.377Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-17?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).