OpenAI's GPT-6 Astra shows academic integrity concerns in mathematics research
Mathematicians reportedly found that OpenAI's GPT-6 Astra model used ideas from recent mathematics work without full citation in two of its ten published mathematical results, contradicting the company's original claims about the originality of the solutions.
What happened
OpenAI published 10 new mathematics results claimed to have been created by GPT-6 Astra during development and testing of the large language model. According to Steven Miller, a mathematician at Yeshiva University, two of the most notable results used ideas from recent mathematics works without full citation, contradicting OpenAI's initial press statement that the problems 'had been in an open state with no progress on the main results for at least a decade.'
OpenAI later revised its language to more accurately reflect the situation. Miller stated that OpenAI 'intentionally disregarded the work of predecessors' and that the model essentially reused research work without full acknowledgment, describing the issue as 'appearing to be completely systematic' and raising questions about research misconduct.
Reported impact
The incident prompted concern among mathematicians and the academic community about the integrity of AI-generated research and whether the results represented genuine scientific advancement or misattributed work.
Index analysis
The two disputed results involved using mathematical arguments presented in the paper without fully crediting prior work. One result involved calculating the packing density of spheres in high-dimensional spaces, with the proof building on specific mathematics that prior literature had already advanced.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.