Generative artificial intelligence, large language models, and agentic rapid-development environments (such as Lovable, Replit, v0, and Cursor) are fundamentally transforming research software. Biomedical researchers increasingly build and publish interactive web applications, analytical dashboards, clinical decision calculators, and bioinformatics pipelines created entirely or substantially through AI generation. However, no standardized reporting guideline exists to disclose how these AI-built scientific web platforms are constructed, verified, and preserved. Existing guidelines (such as CONSORT-AI, STARD-AI, or TRIPOD-AI) govern AI algorithms applied to research datasets, while general AI authorship checklists (such as CHAT-GUIDE) govern text generation in manuscripts. None address the reproducibility, computational accuracy, and data security risks unique to AI-generated software platforms. Unstandardized reporting in this domain introduces four major scientific vulnerabilities:
Model, Platform, and Prompt Provenance: Papers rarely document the underlying foundation model, temperature, platform versions, system prompts, or agentic iteration history used to generate the code.
Algorithmic and Mathematical Verification: Generative AI models frequently introduce subtle mathematical or statistical hallucinations into computational logic. There is currently no reporting checklist requiring authors to verify in-browser or backend calculations against gold-standard scientific packages (such as R Bioconductor or Python SciPy).
Data Governance and Patient Privacy: AI-scaffolded web platforms often introduce undocumented third-party telemetry, cloud database syncing, or external API calls, creating severe privacy risks when handling human or clinical data. Authors must explicitly report whether processing occurs entirely client-side or involves remote servers.
Reproducibility and Software Decay: AI web platforms frequently update underlying dependencies or deprecate hosting runtimes, causing tools to break after publication. Authors must report self-contained source repositories, dependency lockfiles, and long-term archival plans (e.g., Zenodo/Software Heritage).
SWAP-AI will establish an international, consensus-based minimum reporting checklist for any peer-reviewed scientific paper presenting, validating, or deploying web applications and analytical platforms built with AI-assisted or agentic workflows.
Registered: 14 November 2026
Last updated:
Contact: Anas Bedraoui Email: [email protected]