Quantum unclonable encryption: GPT-5.6 Sol Ultra helped close a six-year problem

On July 23, Prabhanjan Anant (UCSB) and Amit Sahai (UCLA) posted a paper on arXiv titled 'Unconditional Unclonable Encryption', closing a six-year challenge in quantum cryptography. The paper's construction and key proof ideas were entirely generated by an agent Codex running on GPT-5.6 Sol Ultra, while humans designed the software harness and verified all claims. The problem is unclonable encryption, where quantum mechanics prevents copying the ciphertext. Classical ciphertexts can be copied; quantum states cannot due to the no-cloning theorem. The goal is to create ciphertext as qubits such that even if the key is later revealed, only one person who holds the original state can decrypt. Security is modeled as a game: an attacker tries to create two useful copies from one quantum ciphertext; security requires that two recipients cannot both guess the message better than random. In 2020, Broadbent and Lord proved a weak version (search security) but the strong version (indistinguishability) remained open for six years. The new proof achieves indistinguishability with advantage exponentially small in n. The proof uses elementary tools (linear algebra, Cauchy-Schwarz inequality) and two key steps: centering an operator and bounding its maximum eigenvalue with a specially constructed filter. Two days later, a second independent proof by Seyoon Raghavan (MIT) using the same GPT-5.6 Sol Ultra also appeared. Both papers are preprints; they claim the AI generated the construction and proof ideas, and the authors take full responsibility.