Paper Pilot Enables Evidence-Traceable Scientific Manuscript Generation with Human Oversight

Large Language Models (LLMs) are increasingly used in scientific workflows for tasks like literature analysis and drafting. However, autonomous AI systems pose a governance challenge, as ideas, methods, and claims can propagate without mandatory human approval or clear traceability to evidence. Paper Pilot addresses this by proposing a human-in-the-loop expert system for evidence-traceable scientific manuscript generation. It adapts the Collaborative Agent Reasoning Engineering (CARE) methodology, incorporating eight approval gates throughout the manuscript development pipeline. The system distinguishes between literature-grounded and artifact-grounded claims, ensuring that all reported numbers and interpretations are traceable to approved evidence. Empirical validation showed that ungated LLM drafters fabricated up to 25% of citations and failed to flag evidence gaps. In contrast, the same models operating under Paper Pilot's evidence-locked rules produced zero fabricated citations and explicitly identified planted gaps. This framework positions LLM-assisted writing as a controlled human-AI decision-support process, rather than fully autonomous authorship, enhancing reliability and accountability in scientific communication.