AI Plagiarism Scandal: Huawei Fights Allegations of Copying Alibaba

Huawei has found itself at the center of a significant controversy in the artificial intelligence sector, pushing back against serious allegations that its recently open-sourced Pangu Pro MoE model is a plagiarized version of a model developed by its domestic rival, Alibaba. The dispute has ignited a firestorm of discussion online and within the tech industry, highlighting the intense competition and scrutiny surrounding AI development in China. The core of the issue revolves around claims that Pangu Pro MoE, which Huawei released in late June as part of a strategic push to build an ecosystem around its proprietary Ascend AI hardware, is not an original creation but rather a "recycled" or "upcycled" product based on Alibaba's Qwen-2.5 14B model.[1][2] These accusations threaten to tarnish Huawei's reputation as a homegrown tech champion and raise broader questions about intellectual property and transparency in the rapidly evolving world of generative AI.

The allegations first surfaced from an entity calling itself HonestAGI, which published a technical paper on the code-sharing platform GitHub.[3] The now-removed report claimed to have found an "extraordinary correlation" of 0.927 between the parameter matrices of Huawei's Pangu Pro MoE and Alibaba's Qwen model.[2][4] HonestAGI's analysis was based on a novel "fingerprinting" technique that examines the standard deviation patterns of a model's attention parameters, which the group argued serve as an intrinsic signature of a model's architecture and training history.[2] The paper contended that this statistical fingerprint is robust enough to persist even after continued training, a method that could otherwise obscure the origins of a copied model.[2] Bolstering their claims, the researchers pointed to other architectural similarities and the discovery of an Alibaba Qwen license file within Pangu's official code repository.[2] The paper concluded that the sheer number of overlapping traits made coincidence highly improbable and suggested potential copyright violation and false claims by Huawei about its investment in training the model from scratch.[2][5]

Huawei's response was swift and unequivocal. The company's Noah's Ark Lab, the research division responsible for the Pangu models, issued a firm denial, stating that the model was independently developed and trained.[4][6] The lab asserted that Pangu Pro MoE was "not based on incremental training of other manufacturers' models" and that it represented "key innovations in architecture design and technical features."[5][3] A cornerstone of Huawei's defense is that the model was built entirely on its own Ascend full-stack AI software and hardware platform, a key part of the company's strategy to create a self-sufficient technology stack in the face of stringent U.S. sanctions.[1][7] While Huawei's statement acknowledged the use of some third-party open-source code, it stressed that its developers strictly adhered to all open-source license requirements and that copyright declarations were clearly marked in the source code files.[7][8] The company framed this as a standard and collaborative practice within the open-source community.[7]

The controversy has been further complicated by unverified online posts. An unconfirmed letter, purportedly from a resigning member of the Pangu development team, began circulating, appearing to corroborate the plagiarism claims by mentioning the copying of codes and watermark removal.[7] Separately, HonestAGI claimed to have received messages from multiple whistleblowers within Huawei who allegedly confirmed the accusations.[2] However, the identity of HonestAGI remains unknown, and the whistleblower claims are unverified.[2][6] This clash comes at a critical time for China's technology sector, which is engaged in a fierce "AI price war" and a race for technological supremacy.[9][10] Companies like Huawei, Alibaba, and DeepSeek are competing intensely to develop foundational models and capture market share, with open-sourcing becoming a key strategy to drive hardware sales and build developer ecosystems.[2][9][11]

The dispute between two of China's most prominent tech giants carries significant implications for the AI industry, both domestically and globally. It casts a spotlight on the increasingly blurred lines between legitimate open-source collaboration and intellectual property theft in the development of large language models.[1] As the technology becomes more complex, proving or disproving such allegations can be technically challenging. The incident also coincides with evolving legal frameworks in China around AI-generated content and intellectual property, which will likely face new tests as disputes like this arise.[9] For Huawei, a company that has positioned itself as a symbol of China's technological independence, the allegations are a direct challenge to its narrative of indigenous innovation.[11][8] The outcome of this controversy could have a lasting impact on corporate reputations, the competitive dynamics of China's AI market, and the level of trust and transparency expected within the global open-source community.