A theoretical computer scientist, citing sources, says AI labs have quietly started probing whether their models can break important cryptographic protocols

AI models may soon be capable of breaking encryption, a core component of secure digital communication.

A theoretical computer scientist, Scott Aaronson, has reported that AI labs are investigating whether their models can compromise cryptographic protocols. This probe is reportedly in response to OpenAI's recent release of numerous mathematical proofs, some of which are so complex that human experts are struggling to understand them. The proofs, generated by OpenAI's latest internal model, have solved long-standing open problems across various fields of mathematics and theoretical computer science. These include the Unique Games Conjecture, L=BPL, and advancements in matrix multiplication and quantum complexity theory. The AI model reportedly solved approximately 5% of the problems it was tested on, with each problem requiring about three hours of compute time. The method of sharing these AI-generated proofs, whether through immediate public release or a more curated approach, is a subject of discussion.

The AI's ability to generate proofs for complex mathematical conjectures suggests a potential leap in AI's reasoning and problem-solving capabilities. If these models can indeed crack cryptographic protocols, it signifies a paradigm shift in cybersecurity, necessitating the development of quantum-resistant or AI-resistant encryption methods. This could render current security infrastructures obsolete, impacting everything from financial transactions to national security.

The speed at which these AI models are advancing, coupled with their potential to disrupt fundamental technologies like cryptography, indicates a rapidly evolving technological landscape. The current debate around how to responsibly disclose and verify these AI-generated breakthroughs highlights the challenges of integrating such powerful tools into scientific and technological progress. The implications extend to the future of research itself, potentially democratizing complex problem-solving while also raising questions about intellectual property and human expertise.