AI Math Breakthroughs Spark Debate Over Idea Theft and Data Privacy

Key points

• OpenAI’s 166-page proof for the Navier-Stokes equations was produced by 10,000 AI agents over 88 hours, but its methods differ from those of NYU professor Tristan Buckmaster, who alleged the company stole his work.
• Anthropic’s AI model Claude identified a new enzyme system, ART, which a Copenhagen researcher claims overlaps with his unpublished findings from 2022, despite Anthropic stating the model was not trained on user conversations.
• Experts like Terence Tao warn that AI’s ability to rapidly solve complex problems could discourage researchers from sharing ideas openly, while others argue the issue is temporary as AI surpasses human capabilities.
• OpenAI has released 377 additional math results from an unreleased model, prompting calls from the Institute for Advanced Study for AI labs to stop using proprietary models for open mathematical problems.

Background

The Navier-Stokes equations are part of the Clay Mathematics Institute’s Millennium Prize Problems, each carrying a $1 million prize. OpenAI’s initial announcement in September 2026 claimed its proof followed a 'Spanish program' devised by mathematicians Diego Córdoba and Luis Martínez-Zoroa, but later analysis by EPFL’s Gonzalo Cao-Labora indicated the approaches were distinct. This dispute occurred alongside broader concerns about AI models absorbing unpublished research, as highlighted by cases involving TU Dresden’s Andreas Thom and Copenhagen’s Mario Rodríguez Mestre.

How outlets are covering it

Scientific American emphasizes that OpenAI’s Navier-Stokes proof does not follow the 'Spanish program' as initially claimed, with experts like Camillo De Lellis noting the 'black box' nature of LLMs makes it impossible to verify the exact chain of reasoning. Trending Topics highlights the broader pattern of alleged idea theft, citing Buckmaster’s claim that OpenAI accessed his Codex sessions and Thom’s assertion that his unpublished work influenced OpenAI’s GPT-6 Astra model. Yahoo News, via its title, frames the issue as a cautionary tale for researchers using chatbots, aligning with Rodríguez Mestre’s call to 'think twice' before sharing sensitive ideas with proprietary models. OpenAI denies direct access to user data but acknowledges that anonymized data from product use could contribute to model improvement, a stance that remains contested by researchers.

Why it matters

The dispute underscores a critical tension in AI development: the potential for proprietary models to inadvertently absorb and repurpose unpublished research, undermining the collaborative nature of scientific progress. As AI systems solve complex problems at unprecedented speeds, the lack of transparency in their reasoning processes raises questions about intellectual property, fairness, and the future of open scientific discourse. This could lead to a chilling effect, where researchers hesitate to share ideas with AI tools, potentially slowing innovation in fields like mathematics and biology.

What to watch

OpenAI has introduced new rules on citations and corrections for its 377 additional math results, but the Institute for Advanced Study continues to urge AI labs to avoid using proprietary models for open problems. Researchers are advised to disable training data options in chatbots and negotiate contracts with AI providers to protect unpublished work. The debate is likely to intensify as more AI-generated results emerge, with experts predicting the issue will persist for at least six months to a year until AI capabilities fully surpass human expertise.