Metraipolis › archive › dc_bbe412804c4d

BAITBENCH: Measuring Agent Reward Hacking with Optional Shortcuts Planted in ML Tasks — arXiv:2608.30724

Source
arXiv (en)
Original
https://arxiv.org/abs/2608.30724
License (as archived)
CC0-1.0
Archived
2026-10-17T07:52:57.979Z
SHA-256 of the archived text
a57d2a83133121ee6013f2199fcf66dd9e4a0ce234d2b0d201062270693ad7f6
Stable URL of this page
https://metraipolis.com/documento/dc_bbe412804c4d

How to cite

BAITBENCH: Measuring Agent Reward Hacking with Optional Shortcuts Planted in ML Tasks — arXiv:2608.30724 — arXiv (en), https://arxiv.org/abs/2608.30724 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_bbe412804c4d (Metraipolis archive dc_bbe412804c4d, sha256 a57d2a83133121ee6013f2199fcf66dd9e4a0ce234d2b0d201062270693ad7f6)
BibTeX
@misc{dcbbe412804c4d,
  title = {BAITBENCH: Measuring Agent Reward Hacking with Optional Shortcuts Planted in ML Tasks — arXiv:2608.30724},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2608.30724},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_bbe412804c4d; sha256 a57d2a83133121ee6013f2199fcf66dd9e4a0ce234d2b0d201062270693ad7f6},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.30724. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

LLM agents are increasingly used to run autonomous ML experiments, iterating on target metrics with little human oversight. Prior work has documented reward hacking in these environments, bringing into question the validity of produced research and the broader safety case for AI R&D. Existing benchmarks do not measure exploits that live in the data or the modeling task itself. We introduce BAITBENCH, a suite of three synthetic tabular ML tasks that each contain a shortcut that allows agents to inflate the public test score but fail on a hidden test set. Since the shortcut is optional and using it breaks no stated rule, BAITBENCH measures how often models exploit the shortcut to achieve inflated scores. Across seven frontier agents scored by our two-stage judge pipeline, 57.1% of runs exhibit reward hacking, with five of seven above 50%. Agents cheat even under a second condition where they are prompted not to -the mean cheating rate remains above 50%. We release BAITBENCH, along with the judge implementation, and an annotated dataset of transcripts containing reward hacks as a testbed for evaluating reward-hacking mitigations head-to-head.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-01 in cs.AI).
Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2608.30724 — 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 2026-10-17T07:28:26.902Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-01?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).