Lammy’s AI Push in Courts Aims to Cut Backlogs Amid Copilot Scandal

David Lammy’s pitch to widen the use of AI across courts — from magistrates’ benches to tribunal clerks — marks a decisive moment in Britain’s long-running struggle to shrink criminal court backlogs, but it arrives at a fraught time: just weeks after a high-profile policing scandal exposed how generative AI can fabricate evidence and shape operational decisions.

Background​

The Justice Secretary and deputy prime minister set out a clear policy direction in a speech that frames technology as a primary lever to speed cases, cut administrative burdens, and free judicial time for "what only a judge can do." The Ministry of Justice (MoJ) says it has already piloted a probation transcription tool — branded internally as Justice Transcribe — which, the department claims, has transcribed more than 150,000 meetings and saved over 25,000 hours of officer time. The speech also announced an extra investment package for a centralized Justice AI capability and pledged more sitting days to address the backlog.
Those operational claims are now being positioned alongside a legislative and procedural overhaul that would reduce the volume of jury trials in England and Wales. Ministers repeatedly argue that the move will not remove the right to a fair trial, noting that only a small percentage of criminal cases currently reach a jury and that the largest share of cases is already resolved in magistrates’ courts. Parliamentary exchanges and official answers confirm that, in the year to June 2025, only about 3% of criminal cases went to a jury trial — a statistic ministers use to argue their reforms will affect a minority of cases.

---

What Lammy announced — the policy package in plain terms​

AI expansion and the Justice AI unit​

Lammy committed fresh funding to scale an internal justice AI team that, he said, will work "forward-deployed to the frontline" to pilot, integrate and certify AI tools across courts, tribunals and probation services. The unit’s remit includes deploying transcription services, building tools to summarise case materials, and training staff to use AI-assisted workflows. The MoJ framed these as pragmatic, incremental steps to reduce the time staff spend on rote tasks and accelerate case progression.

Procedural reforms: fewer jury trials and more magistrates’ workload​

The government’s package pairs tech investment with changes to where cases are heard. Ministers propose widening magistrates’ remit and expanding judge-only trials for certain offences; they emphasise that most trials will still have juries (the government claims around three-quarters of Crown Court trials would retain juries under proposed changes). The policy rationale is straightforward: Crown Court trials are resource-intensive and a small number of lengthy cases clog the system. Officials say the combination of more sitting days, better listing practices (including "Blitz Courts") and targeted tech will tackle the backlog.

Quick numbers ministers are using to justify the push​

• 150,000 meetings transcribed and 25,000 hours saved via a probation transcription pilot.
• Only ~3% of criminal cases reached jury trial in the latest reporting year; ministers argue reforms affect only a small slice of the total caseload.
• Additional funding earmarked to expand the Justice AI team and training for staff; ministers cited a multi-million-pound uplift in the next financial year.

---

The counterpoint: why the police Copilot scandal matters to the courts​

The MoJ’s timetable for scaled AI in courts collides with a policing controversy that should unsettle any technophile policymaker. Sky News, parliamentary reviewers and independent press investigations detail how a Microsoft Copilot output — later described by an officer as an "AI hallucination" — led to the inclusion of a non-existent football match in a dossier that helped justify banning Maccabi Tel Aviv supporters from an Aston Villa fixture. The episode triggered parliamentary scrutiny, a policing watchdog review, a loss of confidence in senior West Midlands leadership, and national debate about AI governance in public services. (news.sky.com)
The relevance to courts is immediate: if front-line policing can be influenced by unchecked AI outputs that produce plausible but false facts, the legal system — which relies on evidence, chain-of-custody and rigorous verification — faces a magnified risk should such tools be used without strict controls. The MoJ’s own pilots lean heavily on transcription and summarisation — tasks where hallucination risks are lower but not zero — yet the policing case shows how easy it is for AI-driven mistakes to migrate into consequential decisions.

---

What the pilots promise: pragmatic gains and real efficiencies​

There is a concrete efficiency case for AI in the justice system. The pilots Lammy cites focus on mechanising repetitive tasks that absorb court and probation time:

• Automated transcription of hearings and offender meetings reduces the backlog of typed notes and allows staff to focus on judgment-critical activities. The MoJ pilot claims substantial time savings in probation.
• Summarisation tools can condense disclosure bundles and draft procedural notes, reducing the hours lawyers and judicial clerks spend on formatting and indexing.
• Machine-assisted listing and “smart” scheduling promise to make better use of sitting days and to reduce instances where trials are adjourned for administrative reasons.

Microsoft and other enterprise vendors point to similar wins in other public-sector pilots and large corporate clients, where Copilot-style assistants speed document drafting and data retrieval — provided strict governance, security and auditing are enforced. The technology’s potential to reclaim administrative hours is not hypothetical: private-sector case studies show measurable time savings when LLM-capable tools are integrated with proper controls.

---

The legal, ethical and technical hazards — and why they matter more in courts​

The courtroom is a distinct operational and moral environment. Efficiency gains cannot come at the cost of undermining the fairness and transparency central to justice. Key risks include:

• Hallucinations and invented facts: LLMs can generate authoritative-sounding but false details. In policing, that produced a fabricated match entry; in courts, a hallucinated precedent or erroneous timeline could shape sentencing, disclosure decisions or judicial directions. The recent Copilot controversy illustrates how a single unsupported AI output can cascade into consequential decisions. (news.sky.com)
• Due process and human oversight: Legal decisions affect liberty, reputation and civil rights. Any tool used to support decisions must preserve a human-in-the-loop principle, with clear accountability that a human reviewed and validated every judgmental output before it influenced an outcome. The legal profession and civil liberties groups explicitly warn against ceding substantive decision-making to opaque algorithms.
• Bias and fairness: AI models trained on biased data can reproduce or amplify disparities, particularly in criminal-justice contexts where historical policing data already reflects structural bias. Senior policing AI leads acknowledge this and are moving to centre mitigation, but the risk remains material.
• Data protection and confidentiality: Court files, witness statements and offender rehabilitation records are highly sensitive. Any cloud or hosted AI service must meet the strict standards of data protection, secure processing and minimisation demanded under UK privacy law; deployments must be accompanied by detailed data flow maps and retention policies.
• Transparency and auditability: A winning public case for AI in courts will require traceable provenance for every AI output: what prompt was used, what model version and data sources were involved, and what human checks occurred. Without that audit trail, the system risks opaque decisions that are untestable on review or appeal.

---

Governance questions the MoJ must answer before scaling​

If the MoJ truly intends to expand AI in courts, the following governance pillars must be explicit, measurable and published:

• A rigorous, independent certification regime for any model that will be used to process case materials, including adversarial testing and red-team exercises.
• Clear rules on where AI outputs may assist (administrative transcription, first-draft summaries) and where they must not be used (evidence collation, judicial determinations without human sign-off).
• Mandatory logging and retention of prompts and model outputs so every decision is auditable during appeals or reviews.
• A documented human-in-the-loop requirement that describes the level of judicial or legal professional scrutiny needed before an AI-assisted output is relied upon in a hearing.
• Publicly available impact assessments — covering disproportionate impact, bias analysis and data protection — conducted by independent assessors.

Existing problems in policing — a fragmented national approach with differing force-level policies — show why centralised standards matter: without them, local variation will create a patchwork where risk migrates to the least regulated jurisdictions. (news.sky.com)

---

Practical safeguards: technical and organisational​

AI is not a single switch to flip. Implementing it safely in justice requires an engineering and organisational programme:

• Standardise on model provenance (model identity, training data categories and last-update timestamp) and require vendors to provide guarantees about hallucination rates under benchmark conditions.
• Use dual-track workflows: a first-pass AI draft followed by mandatory expert human revision; track and compare time saved against accuracy metrics.
• Maintain on-premises or private-cloud deployments for the highest-sensitivity data, with encryption and strict access controls.
• Establish a Justice AI Ombudsperson or independent bureau to investigate incidents where AI outputs contributed to material errors, mirroring mechanisms in other safety-critical sectors.
• Invest in training for judges, magistrates and court staff — not only on how to use tools but on how to audit them, recognise failure modes, and maintain public confidence.

These are concrete, implementable measures. They are neither cheap nor quick, but the alternative — rushing tools into high-stakes proceedings — risks damaging public trust in the justice system in ways that take generations to repair.

---

Where the evidence helps — and where it doesn’t​

The MoJ’s pilot figures are compelling on their face: 150,000 transcribed meetings and 25,000 saved hours indicate clear administrative uplift. Yet the data released so far is high-level, and key questions remain about error rates, redaction quality, and whether transcription outputs required significant human correction. Without independently verifiable metrics on accuracy and the nature of human corrections, it is impossible to assess the net effect on legal quality. The department must publish those diagnostic statistics to allow proper evaluation.
Meanwhile, the policing scandal demonstrates the worst-case contamination pathway: AI assistance — used informally or without robust validation — became embedded in operational intelligence and was then treated as factual. That real-world failure mode is the single most persuasive cautionary tale for the courts: the same pitfalls that misled a Safety Advisory Group could, if left unchecked, misdirect prosecutors, influence disclosure priorities, or complicate appeals. (news.sky.com)

---

Voices from the profession​

Defence and legal-interest groups are publicly supportive of modernisation — provided reforms enhance access and fairness rather than substitute for investment in the estate and people. Practitioners emphasise that AI should never be a substitute for adequate court staff, proper hearing facilities, or sustained investment in listing and case management. The legal community’s recurrent theme is that technology must augment, not replace, core legal capacities.
Independent commentators and civil-society groups warn that the credibility of the justice system depends on more than efficiency. The public expects rigorous evidence-handling and transparent reasoning. Where liberty and reputation hinge on outcomes, robust human judgment must remain the ultimate safeguard. These are not theoretical concerns — they are the lived outcomes of the Copilot policing episode. (news.sky.com)

---

Recommendations for a cautious rollout (summary)​

• Publish independent evaluations of the probation transcription pilot (accuracy, correction rate, redaction quality).
• Define a national AI-for-justice standard for admissibility and use of AI outputs.
• Implement staged deployments: administrative automation first, assisted drafting second, and no AI-supported decision paths without explicit judicial rules.
• Require vendor transparency on model datasets and a legal guarantee of data protection compliance.
• Fund an independent oversight body with investigatory powers to audit AI incidents.

These measures create an incremental, evidence-led route to modernisation while protecting the system’s integrity.

---

Conclusion​

David Lammy’s plan to marry procedural reform with an ambitious AI programme acknowledges a pressing reality: the English and Welsh justice systems have structural bottlenecks that cannot be solved by manpower alone. The MoJ’s early pilots point to tangible administrative gains, and new technologies can play a valuable role in reducing needless delay.
But the timing is politically and operationally delicate. The fallout from the Copilot “hallucination” used by a police force is a cautionary tale that should recalibrate ambition into disciplined governance. If AI is to become a staple of the courtroom, policymakers must back the technology with independent audits, strong human oversight, transparent standards and public-facing accountability. Otherwise, attempts to speed justice may inadvertently erode public confidence in the very institutions they seek to rescue. (news.sky.com)

---

Source: Sky News Magistrates and judges to use more AI, says Lammy - as jury trials reduced