Which Models Are Our Models Built On? Auditing Invisible Dependencies in Modern LLMs
Modern LLM development has become a tangled web of hidden dependencies, where training pipelines recursively rely on upstream models to generate training data, filter datasets, and evaluate outputs. Researchers have introduced ModSleuth, an agentic system that reconstructs these dependency graphs from public sources with verifiable evidence. The work exposes a critical transparency gap in AI development: the full lineage of most production models remains fragmented across inconsistent documentation, making it nearly impossible for researchers or regulators to trace which models influenced which. This matters because hidden dependencies obscure potential bias propagation, complicate reproducibility claims, and create accountability blind spots as the field scales.
Modelwire context
Analyst takeThe deeper issue ModSleuth surfaces isn't just documentation sloppiness: it's that recursive data dependencies mean a single upstream model's biases or failure modes can propagate silently through multiple generations of downstream products, with no current mechanism to trigger a recall or correction.
This connects directly to the Qwen2.5 fine-tuning work covered here ('System Report for CCL25-Eval Task 5'), which illustrates exactly the dynamic ModSleuth is trying to audit: a LoRA-adapted model trained on a curated dataset whose upstream provenance is never examined. The broader pattern across recent coverage is that the field is rapidly layering specialized systems on top of base models (see also the Doc-to-Atom and MoE router work from the same week) without any standardized way to trace what those base models were themselves trained on. ModSleuth is essentially proposing an audit trail for a supply chain that the field has been treating as a black box.
Watch whether any major model provider formally responds to ModSleuth's dependency graphs by updating their model cards or disputing specific lineage claims within the next six months. Silence from named parties would itself be a meaningful signal about how seriously the industry takes provenance accountability.
This analysis is generated by Modelwire’s editorial layer from our archive and the summary above. It is not a substitute for the original reporting. How we write it.
Modelwire Editorial
This synthesis and analysis was prepared by the Modelwire editorial team. We use advanced language models to read, ground, and connect the day’s most significant AI developments, providing original strategic context that helps practitioners and leaders stay ahead of the frontier.
Modelwire summarizes, we don’t republish. The full content lives on arxiv.org. If you’re a publisher and want a different summarization policy for your work, see our takedown page.