GPT-5.6 Generated a Proof That Solved a 6-Year-Old Cryptography Problem

Cryptographers Prabhanjan Ananth of the University of California, Santa Barbara, and Amit Sahai of UCLA posted a paper on arXiv on July 23, 2026, titled "Unconditional Unclonable Encryption." It closes a problem that had stood in front of quantum cryptography for six years. The Codex agent, running on the GPT-5.6 Sol Ultra model, generated the encryption construction and the key ideas of the proof in their entirety.

What exactly did GPT-5.6 do

Via the Codex agent, GPT-5.6 produced not an isolated fragment of calculations but the whole encryption construction and the main ideas of the proof of its security. Ananth and Sahai state this directly in a separate section at the end of the introduction. That paragraph is being discussed as much as the mathematical result itself: the language model independently proposed a research-level construction rather than carrying out a mechanical calculation.

• Publication date: July 23, 2026, arXiv preprint (2607.21551)
• Authors: Prabhanjan Ananth (UC Santa Barbara) and Amit Sahai (UCLA)
• Model: GPT-5.6 Sol Ultra, working through the Codex agent
• The problem had remained open for quantum cryptographers for about six years
• Topic of the paper: "Unconditional Unclonable Encryption"

What is unclonable encryption

Unclonable encryption is a scheme in which the encrypted message physically cannot be copied. The property rests on the quantum no-cloning theorem. The word "unconditional" in the title means the security is proved without computational assumptions: it does not rely on the premise that some problem is hard to break, and holds unconditionally.

The unconditional variant was precisely the stumbling block. According to the paper, a full construction of this kind of encryption remained an open problem for about six years. As Habr puts it, researchers had "one bullet in the chamber": a single working approach, and a second one proved elusive for a long time.

Why it matters

The paper is a notable case of a large language model producing not a numerical answer but a mathematical construction and a research-level proof idea for an open problem. It shifts the role of AI from "assistant for routine tasks" toward co-author of substantive scientific work.

"The construction and the main ideas of the proof were entirely generated by the

Codex agent running on the GPT-5.6 Sol Ultra model," says the section at the end of the introduction of the Ananth and Sahai paper.

According to arXiv, the preprint appeared on July 23, 2026, and is undergoing the open community scrutiny typical of the platform; peer review in specialized journals still lies ahead. But the mere fact that a model proposed a working construction for a problem that specialist groups had struggled with for six years has already made the GPT-5.6 section the focus of discussion.

What this means

The line between "AI calculates" and "AI proves" is shifting: GPT-5.6 proposed the construction and the proof ideas for a problem where the human community had lacked a second approach for six years. If the result survives peer review, it will be one of the first high-profile examples of a language model serving as the author of the substantive part of a scientific paper rather than as a verification tool.

Frequently asked questions

Which AI model generated the proof?

The Codex agent, running on the GPT-5.6 Sol Ultra model, generated the encryption construction and the main ideas of the proof. The paper's authors report this themselves in the section at the end of the introduction.

Who are the authors of the paper and where is it published?

The authors are Prabhanjan Ananth (University of California, Santa Barbara) and Amit Sahai (UCLA). The preprint "Unconditional Unclonable Encryption" was published on arXiv on July 23, 2026 (2607.21551).