Containment Verification: AI Safety Guarantees Independent of Alignment

Royce Moon, Lav R. Varshney

Keywords: AI safety, AI security, agentic safety, agentic security, agentic systems, long-horizon safety, formal verification, Dafny, large language models

TL;DR: We introduce containment verification, a deductive formal verification paradigm proving that agentic frameworks enforce typed boundary policies for every possible AI action, yielding model capability invariant safety at the agent effect boundary.

Abstract: Agentic frameworks are the software layer through which AI agents act in the world. Existing safety methods intervene on the model and therefore remain conditional on unverifiable properties of learned behavior. We introduce containment verification, which locates safety guarantees in the agentic framework itself. Under havoc oracle semantics, the AI is modeled as an unconstrained oracle ranging over the entire typed action space, and the verified containment layer must enforce the boundary policy for every possible AI output. For boundary-enforceable properties, expressed over modeled boundary events, action arguments, and state, we prove a universal guarantee by forward-simulation refinement and mechanize it in Dafny. We instantiate the paradigm by verifying PocketFlow, a minimalist agentic LLM framework, and use an agentic synthesis pipeline to generate the specification, operational model, and refinement proof under an information barrier against tautological specifications. To our knowledge, this is the first deductive formal verification of an agentic framework, and its guarantee is invariant to model capability over the modeled typed action boundary.

Track: Regular Paper (9 pages)

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 222