Nature Cancer, Published online: 10 March 2026; doi:10.1038/s43018-026-01127-0

Kelly et al. assessed an artificial intelligence system for breast cancer screening in retrospective datasets, followed by prospective feasibility evaluation, and report its accuracy, fairness and clinical implementation in multiple workflow settings.

Nature Cancer, Published online: 10 March 2026; doi:10.1038/s43018-026-01128-z

Warren et al. used data from a retrospective cohort of 50,000 women attending breast screening. Arbitration between human and AI decisions was performed in a reader study following normal arbitration workflow. After arbitration, replacing the second human reader with AI in a double-read breast screening workflow was noninferior to two human readers.

arXiv:2603.07980v1 Announce Type: cross Abstract: As language models (LMs) evolve from chat assistants to long-horizon agents capable of multi-step reasoning and tool use, existing benchmarks remain largely confined to structured or exam-style tasks that fall short of real-world professional demands. To this end, we introduce \$OneMillion-Bench \$OneMillion-Bench, a benchmark of 400 expert-curated tasks spanning Law, Finance, Industry, Healthcare, and Natural Science, built to evaluate agents across economically consequential scenarios. Unlike prior work, the benchmark requires retrieving authoritative sources, resolving conflicting evidence, applying domain-specific rules, and making constraint decisions, where correctness depends as much on the reasoning process as the final answer. We adopt a rubric-based evaluation protocol scoring factual accuracy, logical coherence, practical feasibility, and professional compliance, focused on expert-level problems to ensure meaningful differentiation across agents. Together, \$OneMillion-Bench provides a unified testbed for assessing agentic reliability, professional depth, and practical readiness in domain-intensive scenarios.