Reviewing The Leveson Review Part II

Dr Shaun S. Yates questions The Leveson Review Part II’s conception of “efficiency” and its optimism about AI in criminal justice reform.

Date: 16 February 2026

On the 4th Feb 2026, Leveson’s Independent Review Part 2 was released. There is no doubt, this is a difficult read because of its substance and style. I want to comment upon its style before making my central claim: this work marks a worrisome turn in criminal justice reform towards techno-optimism. In making this claim, this short analysis draws upon both volumes of Part II, arguing that (1) Leveson’s conceptualisation of efficiency is unhelpfully messy and (2) Leveson’s vision of AI in the criminal justice process is unjustified in its optimism.

Leveson’s efficiency review is, ironically, not presented in an efficient style: it lacks succinctness, rigorous citation, clear messaging and coherent organisation. Setting the scene, this document that seeks to tackle delays head-on was delayed on arrival – originally scheduled to be delivered the end of year 2025. Adding to the irony, by Leveson’s own admission, many of the points of Part 2 are repeated throughout each of its subsequent chapters; furthermore, many of the points of Part 1 are reiterated in Part 2. Together, Part 1 and 2 amount to well over 1,000 pages and largely consist of unreferenced policy reform argumentation and recommendations (a concession Leveson makes himself). Leveson insists that serious policy thinkers must not engage in a ‘pick-n-mix’ approach to implementing his recommendations; instead, they must “study both parts of the Review in depth” and deploy all recommendations in tandem. Perhaps self-aware of this colossal undertaking, he does somewhat paradoxically state in Part 2 that ‘slavish adherence’ to his recommendations is not required. Lastly, there is Leveson’s choice to organise each volume so that they have multiple sections titled ‘Chapter 3’, ‘Chapter 4’, ‘Chapter 5’, etc. This is simply a confusing editorial choice. Indeed, by way of example, Leveson’s Part 2 Review Volume 1 includes two sections both titled “Chapter 3”; one marking a subsection with another denoting the start of an actual chapter. Stylistic issues such as these are, of course, of lesser importance than the substantive content of Leveson’s work. Nonetheless, for a document of this calibre it’s form should reflect its message.

Onto the substance then, what is efficiency for Leveson? It is a complex picture. He has ‘principles’ of efficiency, ‘key efficiency drivers’ and also provides a single sentence definition, “efficiency is defined as ‘the proportionate and effective use of time and resources to ensure expeditious preparation and fair resolution of criminal cases’”. Problematically, however, when he expands upon these ideas, they become contradictory or even tautological. For example, early on in Part II Volume 1, he states that, “To achieve justice that is timely, fair and proportionate, the courts must also operate efficiently”. Leveson’s thinking here is tautological: timely justice is efficient justice. This matters because the concept of efficiency is, of course, the central concept which grounds his thinking and subsequent policy reform recommendations. The sloppy way Leveson uses the term efficiency obscures the critical trade-offs at the heart of the criminal justice process: speediness is desirable until it degrades due process, standardisation can improve experiences of fairness for court users until it degrades their participation, managerial targets can displace justice-substantiating judgements, etc. What is odd about this is that Leveson repeatedly demonstrates awareness of these important conceptual tensions. This suggests that, despite its volume, the Review’s conceptual framework is not as refined as it could have been, there is some ironing out to be done.

Yet, it is still possible for readers to piece together a coherent understanding of what Leveson means by the term, efficiency. First, Leveson wants the courts to receive a substantial amount of additional funding in a range of targeted areas (as to produce additional court sitting days, investment in the court estate, an expanded administrative workforce, etc.). He also argues that “Money alone will not fix the system, any additional funding risks entrenching the very problems it seeks to resolve” – emphasising that only targeted resource injections are desirable. He believes that strategic financial investments to enhance cross-institutional communication (breaking down information silos) will ultimately result in improved overall resource management; or to use his phraseology, a ‘spend to save’ effect. Improved resource management can then produce hastened case-disposals and mitigate against court delays. In this way, Leveson’s efficiency reforms are centrally about resource management practices that speed-up (and in some cases, fundamentally redesign) conventional court processes and target known delay-causing issues; while expressing a desire to promote justice-substantiating values (such as fairness and effective participation).

As part of this efficiency drive, Leveson has a problematic, borderline utopian vision of the role of artificial intelligence (AI) in the delivery of criminal justice in England and Wales. To take only a few example extracts from his work:

“The pace of technological change means that, within a decade, the environment in which the courts operate may be unrecognisable”

“Responsible use of AI has the potential to reshape operational processes and unlock gains in capacity, efficiency and access to justice”

“I endorse the widespread adoption of these [AI] tools due to the significant efficiency gains they offer.”

Alongside these statements, Leveson does call for safeguards and limits on the integration of AI tools (such as ChatGPT) in the criminal court process. Still, given the scale of what he recommends, the weight he gives to these concerns reads as somewhat performative. By way of example, see his recommendations regarding AI tools and their potential to ‘augment’ human interpreters. When creating these reform recommendations in the 130s range, Leveson acknowledges the current poor performance of AI translation tools. To use the same sources he draws upon, current tools provide an accuracy score of 80% for Spanish, 57% for Vietnamese with other languages receiving an accuracy score as low as 6%. Even in view of this evidence, Leveson maintains that “Based on current progress, AI translation may surpass human interpreting in the near future”, encouraging policy makers to seriously explore the integration of such AI tools for interpreter work. This conclusion is made more surprising by Leveson’s acknowledgement of the wider known problems facing the courts, such as courtrooms’ poor acoustics and the unreliability of simple in-court audio equipment (speakers and microphones). Indeed, in the criminal courts of England and Wales, well-established, simple, proven technologies have been poorly integrated and maintained. In view of this, it is not entirely clear why Leveson is optimistic about the potential of AI, a complex and known-to-be-unreliable technology, to be successfully integrated into the criminal court process. Leveson’s view of AI to revolutionise justice for the better is visionary and optimistic at best, utopian and misjudged at worst.

In conclusion, what is to be made of Leveson’s most recent review? This work is certainly more comprehensive than his 2015 report; undoubtedly, it will be a key text in criminal court policy reform discussions for at least the next decade. While not discussed in great detail in this brief analysis, he does make some recommendations that are sure to please practitioners, socio-legal researchers and social scientists – which includes his call for the courts to receive increased financial support and for the expansion of problem solving courts. Yet, at the heart of this work lies a messy conception of efficiency which incorporates an optimistic vision of AI-enhanced criminal justice; it paradoxically acknowledges the courts’ historic technological failings while encouraging the integrating of complex, emerging technologies. While there are certainly elements within this massive review that should be praised, this central component will also generate justified worry in those concerned for the system’s future.