ShannonProver decomposes cryptographic main theorems into intermediate games and lemmas and generates EasyCrypt proof scripts, cutting expert proof work on ChaChaPoly1305 and MEE-CBC from weeks-to-months to hours

It introduces ShannonProver, an agentic system for cryptographic proofs: given a protocol and its security definitions modeled by a cryptographer, the system decomposes the main theorem into intermediate games and lemmas and constructs EasyCrypt proof scripts for those lemmas. Previously, machine-checked proofs shifted the bottleneck from manual verification to writing the proofs themselves: even when the high-level proof plan is known, turning it into a proof script requires spelling out every detail the plan leaves implicit, which is laborious even for experts; this work assigns that proof-engineering burden to an agent. System description and workflow as stated at the abstract level; implementation details, agent components, and prompting design are not expanded in the loaded text.

It builds a new dataset of EasyCrypt lemmas as an evaluation benchmark, spanning textbook primitives, deployed standardized protocols, and recent NIST proposals, and including expert case studies drawn from a corpus not previously available online. Relative to existing evaluation resources, the benchmark extends coverage to standardized deployed protocols and recent NIST proposals and introduces a previously unavailable expert case corpus. The abstract states the benchmark's coverage and corpus provenance; the number of lemmas, difficulty stratification, and statistics are not given in the loaded text.

On case studies such as ChaChaPoly1305 and MEE-CBC, ShannonProver completes within hours proof developments that historically took experts weeks to months. This moves agent automation from general proof assistance toward end-to-end proof development time scales on real cryptographic protocol case studies. The abstract reports a case-study-level time comparison (hours versus experts' weeks to months), an author-reported case result; per-case timings, success rates, or systematic baseline comparisons are not provided.

The paper suggests a path toward accelerating cryptographic research: as agents automate the proof-engineering burden, cryptographers can iterate more quickly on new constructions, obtain machine-checked assurance earlier, and bring protocols from design to deployment faster. This outlook reframes the value of proof automation from verification efficiency alone to research iteration speed and deployment cadence. A directional judgment offered by the authors on the basis of the case results above, rather than a conclusion independently validated in the loaded text.

The setting this work targets is one in which a cryptographer has already modeled the protocol and its security definitions, with ShannonProver then taking on decomposition of the main theorem into intermediate games and lemmas and construction of EasyCrypt proof scripts. It therefore directly serves formal-cryptography researchers and protocol designers who need machine-checked assurance, especially those working in EasyCrypt workflows. Extensible directions include using proof-engineering automation to iterate faster on new constructions, obtain machine-checked assurance earlier, and shorten the path from protocol design to deployment; the new lemma benchmark also provides a comparable evaluation target for subsequent methods.

The loaded text is abstract-level and contains no figures, per-case timings, success rates, systematic baseline comparisons, or details of the agent's internal components and prompting design, so the stability and failure modes of the method across lemmas of differing difficulty cannot be judged from the available material. The number and difficulty distribution of lemmas among textbook primitives, standardized protocols, and NIST proposals are also not given; readers concerned with benchmark coverage and comparability would need the original dataset description. In addition, the 'hours versus experts' weeks to months' comparison is an author-reported case study, and how the timing is measured and how the expert baseline is defined remain open questions to confirm in the original text.