Copy-Protected Quantum Encryption: GPT-5.6 Helped Prove Six-Year-Old Problem
OpenAI
Cryptographers Prabhanjan Ananth and Amit Sahai proved the security of unconditional unclonable encryption. The proof used AI GPT-5.6 Sol Ultra, which generated the construction and key ideas. Two days later, another group using the same model presented an independent proof of the same problem.
On July 23, a paper by Prabhanjan Anant (University of California, Santa Barbara) and Amit Sahai (UCLA) was posted on arXiv, solving the problem of unconditional non-cloneable encryption that had been open for six years in quantum cryptography. This concerns encryption where the ciphertext is encoded in qubits and exists as a single copy due to the no-cloning theorem: even if the key leaks, only the holder of the original quantum state can read the message. The authors proved strong security (indistinguishability) for a scheme based on eigenstates of Pauli operators, where the key is a classical string of 2n-1 bits and encryption and decryption are performed in linear time. The proof uses elementary tools, including the Cauchy-Schwarz inequality, and gives an exponential bound on the adversary's advantage (2^(-(n+1)/2)). In the section on the use of AI, the authors report that the construction and the main ideas of the proof were generated by the Codex agent on the GPT-5.6 Sol Ultra model; humans created the software harness, verified every claim, and take responsibility for the result. The harness is based on the UCLA Moonshot Harness and a prompt from OpenAI. On July 25, an independent work by Seyeon Raghavan from MIT with the same result, also using GPT-5.6 Sol Ultra, appeared on the Cryptology ePrint Archive. Both proofs are preprints but have already received positive evaluations from experts.
Source: Habr — хаб ИИ —
original