Verified AI The History Behind an Expanding Term and What It Means Today

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

"Verified AI" sounds like a settled category in 2026, but it is not. This paper traces the term in three parts, in the order its history ran. Part I follows verification in artificial intelligence from the verification and validation of expert and knowledge-based systems, the subject of a meeting series that began in 1988, through the 2016 formal-methods formulation of Verified Artificial Intelligence and the assurance frameworks that followed it. Part II shows how that concept shaped the HAIA Ecosystem, from checking AI output against a second platform in 2023 to structured reasoning records, checkpoint governance, hardware-rooted evidence, and governed session records. It then sets out how this work defines Verified AI: AI work whose outputs and decisions leave a record that can be audited and reconstructed, with its evidence rooted in what a human can review and accept, be it in hardware or code. Part III examines the 2026 trend, in which verification language spread across standards, audits, agent identity, and provenance, and it closes with Proof-of-Control, an agent-verification standard released for public comment in September 2026. The standard arrives late to the Verified AI conversation, and by its own terms it stops short of the proof stage, because it places the quality of human oversight outside its evidence. Machines can produce and check evidence, but proof of control, as this paper defines it, requires a named human who verifies that evidence and answers for the decision that follows.