Apple Prose2Policy (P2P): LLM Pipeline Translates Natural-Language Access Policies to Executable OPA Rego — 95.3% Compile Rate, 98.9% Negative-Test Pass Rate

arXiv·medium signalXBlueskyLinkedInCopy link

Apple researchers published Prose2Policy (arXiv 2603.15799), a modular pipeline that converts natural-language access control policies into executable Rego code for Open Policy Agent. The pipeline performs policy detection, component extraction, schema validation, linting, compilation, automatic test generation, and execution. On the ACRE dataset it achieved a 95.3% compile rate with 82.2% positive-test and 98.9% negative-test pass rates — making it practical for Zero Trust and compliance-driven environments where policy-as-code is mandatory.