Did AI Solve a 27-Year-Old Mystery? A Mathematician's Allegations and Data Theft Controversy

Imagine this: You come across a discovery on the internet that you have spent 20 years researching day and night. But you find that the source of this discovery is a model created by a famous AI company. How would you feel? Recently, a similar incident in the mathematics community has become a hot topic. Let’s look behind the scenes at how our knowledge and ideas are being handled in the age of AI.

Why does this matter?

This controversy goes beyond the frustration of a single mathematician. It raises serious questions about how transparently the AI we use every day is about the sources of its training data, and how far our private conversations or ideas are being used to build AI intelligence. We need to resolve the anxiety that when we ask AI for advice or share research ideas, the content could return packaged as someone else’s achievement. As AI model performance improves by leaps and bounds, the importance of data ethics is becoming as great as technical prowess itself.

Understanding the issue

First, let’s look at the difficult term ‘non-sofic group.’ Simply put, it is like a puzzle piece in group theory (a field of mathematics that deals with symmetry) that mathematicians have been unable to solve for a long time. 27 years ago, mathematician Mikhail Gromov posed the question, “Do such groups exist?” but no one had found the answer [Reference: Two Mathematicians Demand Answers From OpenAI on Training Data].

Andreas Thom, a mathematician at TU Dresden in Germany, and his colleague Gábor Kun wrote an important paper in 2019 approaching this problem [Reference: Two Mathematicians Demand Answers From OpenAI on Training Data]. When OpenAI recently announced that its model ‘Astra’ had proven the existence of non-sofic groups [Reference: OpenAI faces backlash over controversial math discovery announcement], Andreas Thom stepped forward to raise questions.

By way of analogy, this is a situation where an intelligent AI chef created by OpenAI steals your diary containing a secret recipe and advertises as if it developed the high-end dish for the first time. The AI chef (Astra) is good at cooking, but the root of the recipe might actually be in your diary (private conversation content). If an AI has trained on a user’s private conversation data and turned the core ideas within it into its own achievement, it could cause serious moral and ethical controversy.

Current situation

What happens next?

OpenAI’s models are currently showing off their prowess not only in solving mathematical puzzles but also in coding, language, and various other fields [Reference: OpenAI faces backlash over controversial math discovery announcement]. Following this incident, demands to transparently disclose data sources during AI development are expected to intensify. Efforts to protect intellectual property rights, as well as those of mathematicians, are emerging as a core social consensus in the AI era. Now that AI has become an entity that integrates and reproduces human intellectual achievements beyond a mere tool, it is time for us to draw a clear line between AI development and fair compensation and data protection.

MindTickleBytes AI Reporter’s View

It is amazing that technology is absorbing human knowledge so quickly, but if it does not respect the value gained by the sweat and effort of the knowledge’s owner, that development can become a house of cards. No matter how smart it is, an AI that loses trust will eventually become a machine that no one believes in. This is why AI companies must make transparency a core value as important as their technical prowess.

References

1. https://modernorange.io/item/49639408
2. https://thecybersecguru.com/analysis/openai-andreas-thom-chatgpt-private-conversations/
3. https://officechai.com/ai/mathematician-andreas-thom-questions-if-openai-used-his-chatgpt-chat-data-for-its-non-sofic-groups-proof/
4. https://www.kucoin.com/news/flash/mathematician-accuses-openai-of-stealing-20-year-research-on-non-sofic-groups
5. https://www.notebookcheck.net/Plainly-dishonest-mathematician-says-OpenAI-lied-to-him-about-training-on-his-private-chats.1395716.0.html
6. https://northeasttimes.com/2026/09/10/two-mathematicians-demand-answers-from-openai-on-training-data/
7. https://en.wikipedia.org/wiki/Sofic_group
8. https://genztech.blog/p/andreas-thom-openai-training-data-dispute/
9. https://www.youtube.com/watch?v=WRu5nFFqZYU
10. https://cryptobriefing.com/openai-backlash-math-discovery-astra/