Metraipolis › archive › dc_eada25e5ffbe

CAVEAT: Towards Robust Computer-Use Agents in Incentive-Misaligned Environments — arXiv:2609.27273

Source: arXiv (en)
Original: https://arxiv.org/abs/2609.27273
License (as archived): CC0-1.0
Archived: 2026-10-08T09:11:10.302Z
SHA-256 of the archived text: 5575f61ff1b40d33c550e34a97791eaf12f251a1774d5baf1c0d4b3b3fad7734
Stable URL of this page: https://metraipolis.com/documento/dc_eada25e5ffbe

How to cite

CAVEAT: Towards Robust Computer-Use Agents in Incentive-Misaligned Environments — arXiv:2609.27273 — arXiv (en), https://arxiv.org/abs/2609.27273 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_eada25e5ffbe (Metraipolis archive dc_eada25e5ffbe, sha256 5575f61ff1b40d33c550e34a97791eaf12f251a1774d5baf1c0d4b3b3fad7734)

BibTeX

@misc{dceada25e5ffbe,
  title = {CAVEAT: Towards Robust Computer-Use Agents in Incentive-Misaligned Environments — arXiv:2609.27273},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.27273},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_eada25e5ffbe; sha256 5575f61ff1b40d33c550e34a97791eaf12f251a1774d5baf1c0d4b3b3fad7734},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.27273. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Computer-use agents (CUAs) increasingly act on behalf of users online. What happens when the environments they operate in have incentives of their own? Online marketplaces, for example, may favor some products over others, steering agents away from the user's objective. Existing CUA benchmarks cover cooperative settings or explicit attacks, but do not test whether agents preserve user objectives when the environment itself has a stake in the outcome. We introduce CAVEAT, a controlled benchmark spanning nine marketplace environments and a taxonomy of eight common steering mechanisms. Across five model families, agents purchase the user-optimal product in 78.6% of matched-control episodes but only 17.3% when steering mechanisms are enabled. Larger models and more reasoning improve robustness, but substantial failures persist. Trajectory analysis and targeted ablations identify three weaknesses in how agents decide: they (1) prematurely narrow the set of alternatives they consider, (2) impose priorities the user never stated, and (3) commit before resolving decision-relevant evidence. Guided by this diagnosis, we develop CAVEAT-Harness, which targets these failures and raises the optimal purchase rate by up to 80.0 percentage points, and show that targeted post-training further improves a smaller open model. These results establish incentive robustness as a distinct challenge for delegated agents, diagnose failure modes, and show how targeted interventions can substantially improve robustness.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-24 in cs.AI).
Categorie arXiv: Artificial Intelligence (cs.AI) ; Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2609.27273 — 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 2026-10-08T08:13:26.478Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-24?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).