Rayne Holland

Papers

Total Citations

H-Index

About

Dr. Rayne Holland is a pioneering researcher at the intersection of artificial intelligence security and open-source model integrity. Their work focuses on the critical vulnerabilities of Large Language Models (LLMs), particularly the emerging threat of supply-chain attacks through plugin ecosystems and low-rank adapter modifications. Holland's most cited paper, "The Philosopher's Stone: Trojaning Plugins of Large Language Models" (2023), with 3 citations, fundamentally exposes how open-source LLMs—despite their democratizing promise—can be covertly weaponized through seemingly benign refinements. This groundbreaking contribution reveals that even cost-efficient adapter tuning, which enables specialized tasks without expensive hardware, creates a dangerous attack surface for trojan insertion. Holland's research has profound implications for AI safety, challenging the assumption that open-source models are inherently more trustworthy. By demonstrating how malicious actors could poison widely-used plugins, they have forced the AI community to reconsider security protocols for model distribution and fine-tuning. Their work serves as both a warning and a roadmap, establishing foundational principles for verifying the integrity of community-contributed model enhancements.