Brief

Researchers have developed CMIP-Forge, an agentic system designed to automate climate science research by integrating scientific literature with climate data analysis. The system utilizes a retrieval-augmented generation (RAG) approach, processing over 6,500 CMIP6 publications and indexing more than 100,000 chunks of information. CMIP-Forge employs a multi-layered Defense-in-Depth architecture, including static code analysis and an adversarial peer-review protocol with independent reviewer models, to ensure the accuracy and methodological soundness of its autonomous workflows. While demonstrating success in complex research tasks, the system also exposed failure modes in its review loop, such as unresolved revisions and the submission of incomplete code. AI