In the world of modern science and technology, a landmark moment has occurred that could revolutionize the perception of the role of artificial intelligence in fundamental research. An employee of the company Anthropic, without specialized mathematical training, launched an experiment that led to unexpected results. The neural network was given the command to attempt to prove one of the most complex mathematical hypotheses, on which humanity has been working for over a century and a half. A prize of one million dollars is still promised for the proof of this hypothesis.
Automation of Scientific Search: 650 Ideas in 36 Hours
The process of the AI working on the task took 1.5 days. During this time, the algorithm demonstrated the ability for autonomous planning and coordination. The model independently coordinated the execution of the task, testing 650 different ideas and approaches to the solution. The scale of computational resources was impressive: the algorithm consumed 31 million source tokens and involved 60 separate sub-agents in the process. This indicates a transition from simple text generation to complex cognitive activity requiring multi-level logic and hypothesis testing.
Verification of Results and Formalization
The key stage was confirming the correctness of the obtained results. Two in-house mathematicians from Anthropic checked the neural network's conclusions and confirmed their correctness. Subsequently, the conclusions were formalized using the Lean interactive theorem prover system. This is a critically important step, as Lean is the standard for verifying mathematical proofs, excluding human error and ensuring absolute logical accuracy.
Context: AI in Mathematics in 2026
This success is a continuation of a series of mathematical discoveries achieved with the help of neural networks in recent years. In the current 2026, algorithms have already resolved several Erdős problems, and the Astra model from OpenAI proved 10 important results. Previously, the Anthropic team also refuted the well-known Jacobi hypothesis. These events indicate that AI has ceased to be just a tool for data processing and has become a full-fledged co-author in the creation of new mathematical knowledge.
Ethical Debates: Authorship and Responsibility
Despite the technical success, the scientific community is divided in opinion. In June, a group of prominent mathematicians signed a declaration expressing concern that AI undermines the value of personal authorship and responsibility for discovery. Critics fear that the automation of proofs could lead to a loss of depth of understanding of processes and a blurring of scientists' responsibility for their conclusions.
Optimistic View: The Future of Mathematics
At the same time, many experts, including Fields Medal laureate Timothy Gowers, believe that the involvement of AI can change mathematics for the better. They believe that neural networks are capable of finding connections and patterns inaccessible to the human mind, which will open new horizons for science. Anthropic's breakthrough confirms that AI is capable not only of solving assigned tasks but also of finding unexpected approaches to solving problems that seemed insoluble.
