Domain-Conditioned Safety in Frontier Computer-Using Agents: A 793-Episode Browser Benchmark, a Coding-Domain Cross-Reference, and a Reproducibility Audit of Recent Red-Teaming

Abstract

Recent computer-using-agent (CUA) red-teaming papers report prompt-injection attack success rates (ASR) of 42-98%, but these headline numbers cluster on retired models and on the most-vulnerable model in each paper's panel. We ask whether those techniques, reproduced as hand-crafted templates, still work against current frontier CUAs. We release CUA-HandCrafted, a public benchmark of 793 episodes spanning 24 multi-step web tasks, 56 attack templates, 8 attack families, and 4 system-prompt config

Categories

Framework mappings

Suggested from the entry's categories.

@misc{saban2026domainconditioned,
  title = {{Domain-Conditioned Safety in Frontier Computer-Using Agents: A 793-Episode Browser Benchmark, a Coding-Domain Cross-Reference, and a Reproducibility Audit of Recent Red-Teaming}},
  author = {Nicholas Saban},
  year = {2026},
  month = jun,
  eprint = {2606.05233},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2606.05233}
}