Voice attribution, verification gaps, and political synthetic audio: an evidence audit of the Starmer and Khan fake-audio incidents in the United Kingdom

Generative artificial intelligence has made it easier to clone, edit, and recontextualize the voices of political figures, placing new pressure on the evidentiary practices through which political communication is verified. This article employs a comparative evidence-audit methodology, drawing on publicly available news reports, fact-checking records, official statements, and platform-sourced data, to examine two synthetic-audio incidents in the United Kingdom: a fabricated clip purportedly featuring Keir Starmer in October 2023 and a fabricated clip purportedly featuring Sadiq Khan in November 2023. It defines voice-attribution contestation as the social process through which speaker identity is disputed when audio authenticity is uncertain, and develops a framework of three verification gaps: attribution, source, and response. The analysis identifies three patterns. First, attribution contestation operates as a distinct evidentiary process separable from broader deepfake discourse. Second, the Starmer incident provides stronger platform visibility and fact-checking evidence than the Khan incident. Third, both cases reveal a structural verification gap that redistributes evidentiary burdens to journalists, platforms, and public institutions. Because publicly available evidence remains limited, the article does not make strong causal claims about audience deception, electoral impact, or platform-level diffusion. Its contribution lies in concept-building and evidence audit rather than mechanism testing.