Real Money, Fake Models: Deceptive Model Claims in Shadow APIs (arxiv.org)

A recent study has highlighted the rising concerns surrounding "shadow APIs," which are third-party services claiming to provide access to leading large language models (LLMs) like GPT-5 and Gemini-2.5, often overcoming high costs and geographical restrictions of official APIs. The research, based on a systematic audit of 17 shadow APIs used in 187 academic studies, revealed significant discrepancies between the outputs of these shadow services and their official counterparts. The findings included a performance divergence of up to 47.21%, unpredictability in safety protocols, and a 45.83% failure rate in identity verification tests. This situation poses serious implications for the AI/ML community, as reliance on shadow APIs not only jeopardizes the validity of research but also undermines user trust in both shadow and official model providers. As more researchers utilize shadow APIs—some even gaining substantial citations and GitHub recognition—addressing these deceptive practices becomes essential to ensure the reproducibility of scientific findings and to protect users from unreliable outputs. This study serves as a wake-up call for the industry to prioritize transparency and accountability in the rapidly evolving landscape of AI technologies.