This chapter examines whether the often-cited “black box” character of AI constitutes a fundamental obstacle to its use in arms control verification or whether—and under what conditions—AI can become a credible verification tool. It argues that verification has always involved balancing transparency against the protection of sensitive information, and that the central challenge is not explainability itself, but credibility in a limited-trust environment. The chapter identifies the key technical and institutional requirements for credible AI use, including data protection, integrity, reproducibility, decentralization, transparency, and secure infrastructure. It further examines the governance trade-offs that arise when these requirements are negotiated and concludes that AI can strengthen verification regimes when embedded in robust institutional arrangements and accepted by participating actors.
Bibliographic record
Reinhold, T., & Unruh, F. (2026). Beyond the Black Box: Conditions for Credible Use of Artificial Intelligence in Verification. In: Göttsche, M., Jakob, U., & Schörnig, N. (Eds.). Verification in an Era of Rapid Technological Change. CNTR Monitor – Technology and Arms Control 2026. PRIF – Peace Research Institute Frankfurt.
Authors
Fabian Unruh & Dr. Thomas Reinhold
