Effect-Transparent Governance for AI Workflow Architectures: Semantic Preservation, Expressive Minimality, and Decidability Boundaries
arXiv:2605.01030v3 Announce Type: replace Abstract: We present a machine-checked formalization of structurally governed AI workflow architectures and prove that effect-level governance can be imposed without reducing internal computational expressivity. Using Interaction Trees in Rocq 8.19, we define a governance operator G that mediates all effectful directives, including memory access, external calls, and oracle (LLM) queries. Our development compiles with 0 admitted lemmas and consists of 36 modules, ~12,000 lines of Rocq, and 454 theorems. We establishseven properties: (P1) governed Turing
The increasing complexity and potential autonomy of AI systems, particularly LLMs, necessitate robust governance frameworks for safety and reliability, driving research into formal verification methods.
This research details a machine-checked formalization for governing AI workflows, ensuring that critical AI systems can be designed to be both expressive and auditable, a key step toward trustworthy AI deployment.
The ability to formally prove that AI systems can retain full computational expressivity while adhering to stringent governance rules changes the landscape for secure and compliant AI development.
- · AI developers
- · Auditors and regulators
- · High-stakes industries (e.g., defense, finance)
- · Formal verification tool vendors
- · Developers of ad-hoc AI governance solutions
- · Organizations prioritizing speed over verifiable safety
More secure and verifiable AI systems become feasible for deployment in critical applications.
Increased trust in AI leads to broader adoption across regulated industries and government sectors, potentially accelerating AI integration.
The establishment of formal governance proofs could become a de-facto standard or regulatory requirement for high-impact AI, shaping future development paradigms.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.AI