OpenAI's Astra AI Model Accused of Plagiarism in Mathematical Breakthroughs
Several mathematicians have accused OpenAI of academic misconduct, claiming its next-generation AI model Astra failed to credit existing research in a recently published 249-page paper detailing mathematical breakthroughs. Specifically, researchers pointed out that Astra's findings on sphere packing and group theory closely mirrored existing papers from 2016 and 2019 without proper attribution. This controversy highlights the emerging ethical and academic standards for AI-generated scientific discoveries, emphasizing that AI systems must be held to the same rigorous citation standards as human researchers. As AI increasingly contributes to high-level scientific research, establishing clear guidelines for attribution and intellectual property becomes critical. The allegations focus on two main areas: a sphere packing argument in dimensions of 1,000 or higher that was first published by Steven Miller in 2016, and a group theory proof regarding sofic groups that combined ideas from 2016 and 2019 papers. In response, OpenAI stated they stand by the correctness of the results and plan to issue a minor update to the paper to address the citation issues.
## BACKGROUND
The sphere packing problem in geometry seeks the densest arrangement of non-overlapping spheres in a given space, which becomes highly complex in higher dimensions. In group theory, a sofic group is a discrete group that can be approximated by finite symmetric groups, and proving whether certain groups lack this property (non-sofic groups) has been a long-standing mathematical challenge.