OpenAI and Anthropic Claim Major AI Breakthroughs in Mathematics and Computer Science.

OpenAI and Anthropic have drawn international attention after reporting that their advanced artificial intelligence models independently produced work related to a major longstanding problem in theoretical computer science. The development has intensified competition between the two companies and raised new questions about how artificial intelligence could transform scientific research. OpenAI announced several research results involving mathematics, computer science and theoretical physics, while Anthropic reported that its Claude Opus 5.5 model had independently arrived at a result concerning the Exponential Time Hypothesis. This hypothesis has influenced computer scientists’ understanding of how difficult certain computational problems are for decades. If the claims withstand independent verification, the results could represent an important milestone in the use of AI to investigate complex scientific questions.

The Exponential Time Hypothesis, commonly known as ETH, is an assumption in theoretical computer science that certain computational problems cannot be solved substantially faster than existing exponential-time approaches. It has helped researchers assess the difficulty of problems involving Boolean satisfiability and has supported numerous studies in computational complexity. OpenAI’s reported work challenges this assumption by presenting an approach that it says can solve a relevant problem more efficiently than previously expected. Anthropic’s announcement that its own model reached a related result independently has added another dimension to the story. However, these claims should not be interpreted as a final scientific consensus. Independent experts still need to examine the mathematical arguments, verify the proofs and determine whether the results fully support the conclusions being drawn.

The significance of these developments extends beyond competition between AI companies. Mathematics and theoretical computer science provide foundations for cryptography, software development, optimisation and numerous technologies used in modern society. Discovering more efficient ways to solve computational problems could eventually influence how researchers approach difficult tasks across these fields. Nevertheless, a theoretical breakthrough does not automatically translate into faster applications or immediate changes to existing technologies. Researchers must determine whether an algorithm works under practical conditions, whether its performance advantages hold for realistic problems and whether the findings can be independently reproduced. In particular, any implications for cryptography would require careful analysis before concluding that widely used security systems have become vulnerable.

The announcements also highlight a broader shift in artificial intelligence, from systems that primarily answer questions or generate content towards models capable of assisting with advanced research. AI models are increasingly being evaluated on their ability to develop algorithms, investigate mathematical conjectures, analyse scientific literature and support researchers in testing new ideas. If these capabilities continue improving, AI could become a more important research partner, helping scientists explore possibilities that might otherwise require considerable time and effort. However, human oversight remains essential. AI-generated proofs can contain subtle errors, and convincing explanations are not substitutes for rigorous verification. Collaboration between AI developers, academic researchers and independent reviewers will therefore be critical to establishing which claimed breakthroughs are genuinely reliable.

Ultimately, the reported achievements from OpenAI and Anthropic illustrate both the promise and the uncertainty surrounding advanced AI research. Their competing announcements suggest that leading AI systems may be approaching a stage where they can contribute meaningfully to difficult theoretical problems, rather than simply summarising existing knowledge. Yet the scientific value of these developments will depend on the quality of the evidence, the conclusions reached by independent experts and the practical consequences of the discoveries. For the technology industry, the race is no longer only about building faster chatbots or more capable coding assistants; it is increasingly about demonstrating that AI can generate original, verifiable knowledge. The coming months of scientific scrutiny will help determine whether these claims mark a genuine turning point in computer science or an early step towards future breakthroughs.