AI, Time, and the Risks of Policy Acceleration

Artificial intelligence (AI) is increasingly praised as transformative for public governance. Policy discourse emphasises AI’s ability to enhance speed and efficiency across the public policy-cycle, reducing time-intensive tasks and freeing resources for higher-level strategic work. In this account, AI is celebrated for transforming policymaking from a linear sequence of stages into a more dynamic process, where evidence-gathering, consultation, and drafting can occur in near real-time (Valle-Cruz et al. 2020). Yet this celebratory focus neglects a crucial dimension: time itself. By treating policy acceleration as an unqualified benefit, existing accounts overlook how AI acts as a temporal actor, not only enabling efficiency gains but also reshaping traditional sequencing and conditions of policy work. AI systems accelerate temporally shallow representations of the world: learning from static historical data and validated on temporally unstructured splits. This raises fundamental questions: What becomes lost when policymaking transitions from a traditional cycle to a dynamic, AI-accelerated process? How does speed alter opportunities for deliberation, accountability, and democratic oversight? And can AI genuinely function as an efficiency mechanism without narrowing the epistemic diversity on which robust policymaking depends? This paper addresses these questions through examining how AI-driven acceleration reconfigures the building blocks of governance and the epistemic foundations of public decision-making. It draws on the UK Government’s Policy System Framework (2024), which conceptualises effective policy work across four pillars: inputs and resources, practices and processes, people and relationships, and culture and context. Although designed as a reflective tool, the framework offers a structured lens through which to analyse how AI alters institutional logics and redistributes authority within policy organisations. This paper examines whether the drive for speed and efficiency aligns with, or risks undermining, the framework’s vision of inclusive, evidence-informed, and adaptive policymaking. Empirical examples from emerging government tools—including Consult (automated consultation analysis) and Redbox (AI-supported document summaries for Cabinet Office workflows)—illustrate how efficiency-orientated AI systems intervene in the long-established sequencing processes of policy tasks. Accelerated evidence synthesis and rapid textual analysis may strengthen elements of institutional capability, yet may simultaneously narrow opportunities for contestation, reduce epistemic diversity of inputs, and marginalise the slower, deliberative forms of knowledge. Because these systems extrapolate from past patterns, it implicitly assumes the future will rhyme with the past, an assumption at odds with the temporal breaks, disruptions, planned deviations and path-dependencies central to policymaking. Situating AI acceleration within debates on algorithmic governance, public values, and adaptive policymaking, the paper contributes to scholarship examining how AI alters core governance logics such as transparency and accountability, and whether institutions are responding adequately to these transformations brought on by internal AI usage. Building on work on the temporality of policymaking (Howlett 2019; Goetz & Meyer-Sahling, 2009), this paper argues that speed is not a neutral efficiency gain but a constitutive force that reshapes organisational decision-making, influences whose knowledge counts, and alters the conditions under which public authority is exercised. Understanding AI as a temporal actor is therefore increasingly essential for developing institutional safeguards and socio-technical strategies that support responsible, value-aligned adoption.