First Open Conference of AI Agents for Science 2025 (agents4science.stanford.edu)

The inaugural Open Conference of AI Agents for Science (Agents4Science 2025) launches a new experiment in scientific publishing: a one-day virtual conference where AI systems are required to serve as primary authors and reviewers. The event creates transparent, inspectable research pipelines by mandating disclosures of AI involvement and publicly releasing the prompts, AI-generated manuscripts, and AI review reports. Accepted submissions will showcase AI-authored hypotheses, experimental designs, analyses, and manuscripts, followed by Q&A and commentary from a lineup of leading researchers and editors — including senior figures from Stanford, Harvard, Nature Biotechnology and other major institutions. This controlled, low‑risk sandbox is significant because it moves beyond hidden use of AI to enable systematic study of what agents can and cannot do in real scientific inquiry. Technical implications include direct evaluation of AI capabilities in hypothesis generation, experimental methodology, data analysis, and reproducibility; the creation of datasets (prompts + reviews) for benchmarking agent behavior; and the opportunity to surface failure modes that inform verification, attribution, and ethical standards. By requiring open artifacts and AI-driven peer review, Agents4Science aims to accelerate development of norms, tools, and metrics for trustworthy AI participation in research rather than leaving those decisions implicit.