The UK Criminal Justice System Is One Technological Breakthrough Away from Catastrophic Collapse

I build intelligence systems for a living. I design architectures that take fragmented signals — conversations, prospect data, market patterns — and synthesise them into something actionable. My job, in essence, is to make hidden things visible.

Recently, I started wondering what would happen if I pointed the same lens at one of the systems that holds society together. Not to sell anything. Not to prove a political point. Just to see what the numbers actually say when you model them honestly.

I chose the criminal justice system in England and Wales. I chose it because it is a complex system with publicly available data, it is already under extraordinary strain, and it is about to be hit by forces that nobody in government is modelling for.

What I found was worse than I expected. Not because the system is underfunded, though it is. Not because the politics are broken, though they are. But because the system is architecturally incapable of surviving a level of technological change that is not hypothetical. It is already underway.

I have a daughter. She is a few months old. I started this analysis as an intellectual exercise. I am finishing it as a father who wants to understand the shape of the country she is going to grow up in.

Part 1: The System Today

Let me start with the numbers as they stand right now, before any external shock.

The Funnel

The Crime Survey for England and Wales estimates approximately 9.3 million incidents of headline crime per year. This figure captures theft, robbery, criminal damage, fraud, computer misuse, and violence. It does not include many categories of crime that go unreported or that the survey cannot capture well, including most drug offences, many sexual offences, and crimes against businesses and children.

Of those 9.3 million estimated incidents, approximately 6.6 million are recorded by the police. The gap between the two figures tells you something important: roughly 30% of crime that victims themselves report experiencing never makes it into the official police record.

From there, the funnel narrows sharply. 1.58 million individuals were formally dealt with by the criminal justice system in the year ending June 2025. Of those, 1.36 million defendants were directed to appear at magistrates' courts. 1.20 million offenders were sentenced. And of those, 83,700 received an immediate custodial sentence.

The system is not designed to imprison everyone. It was never intended to. But the scale of the gap between what happens and what the system responds to is the first thing you need to understand, because it is the gap that technology is about to close.

The Bathtub

The prison population of England and Wales currently stands at approximately 87,000 to 88,000 people. This is not the number of people sentenced in a year. It is the total number of people in the system at any given time — a stock figure that accumulates from multiple years of sentencing.

Think of it as a bathtub. Roughly 83,700 people flow in per year. The drain — people completing their sentences and being released — operates slowly because average sentence lengths have been increasing. The average custodial sentence is now 19.5 months, and for indictable offences it is 21.5 months. Average sentence length has increased by 45% since 2010. People are staying longer. The tub fills up.

The system has a designed capacity of approximately 80,000 places that meet the prison service's own standard of decent accommodation. It is currently holding 88,000 people. Seventy-two percent of prisons are overcrowded. Some, like HMP Leeds, are running at 172% of their certified normal capacity. The male prisoner estate has recorded occupancy rates as high as 99.7%, against a safe operational threshold of 95%.

Nearly half of all prisons in England and Wales were judged to have concerning or seriously concerning performance in the most recent assessment. Seven people were killed inside prisons in 2024 — compared with nine in the previous five years combined. Violence against staff is up 19%. Prisoner-on-prisoner assaults are up 14%.

The Courts

The crisis does not begin in the prisons. It begins in the courts.

The Crown Court backlog in England and Wales currently stands at approximately 80,000 open cases. This has nearly doubled since before the pandemic. Courts are processing fewer cases per day than they were in 2019, despite average case complexity falling. The median time from offence to completion at the Crown Court is 338 days — more than eleven months.

The government's own projections, which assume no external shocks, show the Crown Court backlog reaching 130,000 by 2030 and exceeding 200,000 by 2035. The Justice Secretary has described the situation as "absolutely dire." In response, the government has introduced "Swift Courts" where cases with a likely sentence of three years or less are heard by a judge alone, without a jury. More than 3,200 lawyers signed a letter opposing this measure.

The courts are the bottleneck. They are the narrowest point in the entire system. Every case that enters the system must pass through them. And they are already failing at current volumes.

The Revolving Door

The government's primary response to the prison capacity crisis has been early release. Prisoners are being released before their sentences are complete, under supervision, to free up space.

The results have been predictable. In the year ending September 2024, there were nearly 32,500 admissions to prison following recall — people who had been released and were brought back. This represents a 27% increase on the previous year. In a single quarter, there were 9,523 recall admissions, up 39% on the same period the previous year. Both figures are record highs.

Three quarters of those recalls were for non-compliance rather than new offending. The system released people who were not ready, they breached their conditions, and they came straight back — consuming the capacity that the early release was supposed to create.

In the year ending March 2025, 262 prisoners were released in error — an increase of 128% on the previous year. The Ministry of Justice attributes this to staff having to process a range of operational and legislative changes that the system was not designed to absorb.

The cost of holding one prisoner for a year is £53,800. Multiply that by 88,000 and the annual bill exceeds £4.7 billion. That is what the country is paying for a system that, by its own measures, is overcrowded, unsafe, and failing to rehabilitate the people it holds.

Part 2: The Variables Nobody Is Modelling

The projections cited above — 100,000 prisoners by 2030, 200,000 court cases by 2035 — are linear extrapolations. They assume that crime rates, detection rates, sentencing patterns, and economic conditions remain roughly as they are today. They assume continuity.

I do not believe continuity is a reasonable assumption. Here is why.

Variable 1: The Visibility Paradox

This is the variable that I believe matters more than any other, and the one that I have not seen discussed in the policy literature.

The criminal justice system is currently calibrated for a world in which most crime goes undetected. The Crime Survey estimates 9.3 million incidents per year. The police record 6.6 million. Approximately 1.58 million people are formally dealt with. The system has been designed, funded, staffed, and built around the assumption that the vast majority of crime never enters the pipeline.

I work with AI-driven intelligence systems every day. I know what happens when you increase the resolution of a detection system. You do not just see more of the same. You see things that were always there but invisible. The pattern existed. You just could not resolve it before. And every time you increase resolution, the volume of actionable signal does not increase linearly. It explodes.

Now apply that to law enforcement. AI-powered surveillance is already being deployed: facial recognition, automated number plate readers, AI-assisted report writing on body-worn cameras, real-time analysis of cell phone footage, predictive policing algorithms, and financial transaction monitoring. Each of these tools, individually, increases the detection rate for specific categories of crime. Combined, they begin to change the fundamental ratio between crime committed and crime detected.

And here is what makes this a true paradox rather than merely a problem. Once you can see a crime, you cannot choose to unsee it. If the technology exists to detect an offence and the state chooses not to act, that is a policy decision. That is the government deciding which laws to enforce and which to ignore. That is not justice. That is rationing dressed up as discretion.

The Visibility Paradox does not require speculation about future technology. It requires only a modest improvement in what already exists.

Variable 2: Labour Market Displacement

The UK workforce comprises approximately 33 million people. Conservative estimates from the OECD and major consultancies suggest that 10 to 30% of UK jobs are highly exposed to automation in the coming years. Even at the lower end, that is 3.3 million people under significant economic pressure.

The displacement will not be evenly distributed. It will hit the middle of the skill distribution hardest — the administrative, clerical, and mid-level knowledge work that is most susceptible to AI automation. This creates a specific demographic: people with enough education to feel the injustice of their situation, but not enough adaptability or resources to pivot quickly.

Historical data is clear about what happens when large populations experience sustained economic pressure. Property crime increases. Domestic violence increases. Fraud increases. Public order offences increase. Drug and alcohol dependency increases. Mental health deteriorates, which in turn increases the likelihood of behaviours that intersect with the criminal justice system.

The government's AI strategy is focused on growth and competitiveness. There is no serious published plan for managing the transition for the millions of people whose roles will be degraded or eliminated.

Variable 3: Intelligent Crime

This is the variable that should concern policymakers most and is discussed least.

The current criminal justice system is calibrated for opportunistic and desperate crime. Theft driven by addiction. Violence driven by impulse. Fraud executed clumsily. The detection rate is low partly because enforcement tools are limited, but also because most crime is not sophisticated.

Labour displacement will change this. When mid-level knowledge workers — people who understand systems, financial structures, data architectures, corporate processes, and regulatory gaps — face sustained economic pressure, they do not commit impulsive crime. They build.

They build fraud systems that are architecturally sound. They find regulatory arbitrage. They use AI tools to automate scams at scale. They understand how to move money, manipulate data, and create synthetic identities. They know where the gaps are because they used to work in the gaps.

This category of crime takes longer to investigate, longer to prosecute, and consumes dramatically more court time per case than the offences the system currently handles. Even a modest increase in the volume of complex economic crime would place disproportionate pressure on an already failing court system.

Variable 4: The Compounding Squeeze

The variables above do not operate in isolation. They compound.

The cost of living continues to apply pressure at the household level. Food prices remain elevated. Energy costs, water infrastructure, and housing are increasingly dominated by institutional and foreign ownership that prioritises extraction over affordability. Consumer credit carries punitive interest rates, and the credit scoring system penalises those who fall behind with higher charges, creating a downward spiral.

Each of these pressures independently correlates with increased contact with the criminal justice system. Combined, they create the conditions for a sustained increase in the volume of people entering the top of the funnel — at exactly the moment when AI-driven detection is increasing the system's ability to see them.

Part 3: The Cascade

What happens when these variables interact? The answer is a cascade — a sequential failure where each system that exceeds its capacity creates additional pressure on the next.

Stage 1: Detection Outpaces Processing

AI tools increase the volume of detected crime. Police forces, under pressure to justify their adoption of new technology, feed more cases into the system. The Crown Prosecution Service, already stretched, faces a step change in incoming volume. Charging decisions become slower. The backlog between arrest and charge grows.

Stage 2: Courts Seize

The Crown Court backlog, already at 80,000 and projected to reach 130,000 by 2030 on current trends, accelerates past projections. Trials are listed for 2030, then 2031. Victims withdraw. Witnesses forget. Evidence degrades. The conviction rate falls not because defendants are innocent, but because the system is too slow to deliver justice.

The government extends judge-only trials to more categories. Due process is traded for throughput. Constitutional safeguards are quietly eroded in the name of efficiency.

Stage 3: Prisons Overflow

The prison population, already at 88,000 against a decent capacity of 80,000, pushes toward 100,000 and beyond. Early release schemes are expanded. Sentence lengths are cut administratively. People are released without rehabilitation, without housing, without support. The recall rate, already at record highs, accelerates. The revolving door spins faster, consuming the capacity it was meant to create.

Stage 4: Triage Becomes Policy

The system can no longer process the volume. Prosecutorial discretion — the CPS deciding which cases to pursue — becomes the de facto sentencing mechanism. Entire categories of offence are quietly deprioritised. Police stop arresting for crimes they know will not be prosecuted. An unwritten hierarchy emerges: which crimes matter and which do not.

This is not a policy decision made in Parliament. It is an operational decision made in crisis, without public debate, without democratic accountability.

Stage 5: Surveillance Replaces Incarceration

If the system cannot imprison people, it must monitor them. GPS tags, digital curfews, algorithmic risk scoring, and real-time location tracking become the default consequence for all but the most serious offences. The technology exists today. The political resistance to deploying it at scale evaporates the moment the alternative is visible disorder on the streets.

The state does not choose a surveillance model ideologically. It arrives at one operationally, because every other option has failed.

The Timeline

This cascade does not require a catastrophe. It does not require a recession, a pandemic, or a social crisis. It requires only that the trends already in motion — AI detection capability, labour market pressure, compounding economic strain, and court system degradation — continue on their current trajectories and interact.

The early stages of the cascade are already visible. Prisoners are being held in police cells. Court appearances are being delayed because there is no room. Prisoners are being released in error because staff cannot manage the complexity. Seven people were killed in prison last year. These are not warning signs. These are symptoms of a system that has already exceeded its design limits.

On the government's own projections, which assume no external shocks, the court backlog reaches crisis levels within five years and the prison population exceeds any plausible capacity scenario within the same timeframe.

Introduce any one of the variables described above and the timeline shortens. Introduce all of them simultaneously, which is what is actually happening, and the cascade does not begin in five years. It is already underway.

Part 4: What Better Looks Like

I am not a policy expert. I am a systems architect. I design structures that process information at scale, route decisions efficiently, and produce outcomes that are proportionate to inputs. When I look at the criminal justice system, I see an architecture that was designed for a different era, operating under assumptions that are no longer valid, consuming enormous resources to produce outcomes that by its own metrics are failing.

The question I would ask if this were a product architecture is: if we were building this from scratch today, with everything we know about data, AI, behavioural systems, and institutional design, would we build anything that looks like this?

The answer is obviously no.

A system designed today would start from different principles. It would recognise that deterrence depends on certainty and speed, not severity and duration — the evidence on this is overwhelming and has been for decades. It would build consequence structures that are proportionate, swift, and functional at every level, rather than funnelling everything toward a single mechanism — incarceration — that is expensive, overcrowded, and demonstrably failing to reduce reoffending.

It would use the same intelligence and data infrastructure that is increasing crime visibility to manage consequences more effectively. Tiered monitoring systems, graduated responses, structured reintegration pathways, and community-based alternatives that are genuinely resourced rather than presented as afterthoughts.

It would address the economic conditions that drive the majority of crime into the system in the first place, recognising that a justice system cannot compensate for failures in housing, employment, mental health, and social support.

And it would be honest about what it can and cannot do. The current system pretends to be comprehensive while converting less than one percent of crime into custodial consequences. A better system would acknowledge its capacity limits and design within them, rather than promising justice it cannot deliver.

I have ideas about what that architecture looks like in detail. I will write about them separately. For now, what matters is the diagnosis: the system as it stands is not merely underfunded or poorly managed. It is structurally incapable of absorbing the forces that are about to hit it. And the people responsible for preparing for those forces are modelling a future that assumes none of them will arrive.

Sources

• Office for National Statistics, Crime in England and Wales: Year Ending September 2025
• Office for National Statistics, Crime in England and Wales: Year Ending March 2025
• Ministry of Justice, Criminal Justice Statistics Quarterly: June 2025
• Ministry of Justice, Criminal Court Statistics Quarterly: April to June 2025
• Ministry of Justice, Prison Population Statistics, 2025
• Ministry of Justice, Prison Population Projections: 2024 to 2029
• House of Commons Library, Prison Population Statistics, February 2026
• Public Accounts Committee, Prison Estate Capacity Report, March 2025
• Prison Reform Trust, Bromley Briefings Prison Factfile, February 2026
• Prison Officers' Association, UK Prison Crisis Report, 2025
• Howard League for Penal Reform, Why Are Prisons Overcrowded?, 2025
• Institute for Government, Performance Tracker 2025: Criminal Justice System
• National Audit Office, Reducing the Backlog in the Crown Court, March 2025
• Sentencing Academy, Sentencing Trends 2005–2024, February 2026
• Law Gazette, Crown Court Backlog Projections, February 2026
• GOV.UK, Swift and Fair Plan for Victims, 2025

This is Stress Test 001 in a series examining public systems under pressure. Each piece takes a system, maps its current capacity, introduces the variables that are about to change, and models what happens when they compound. Nothing on this site is for sale. The analysis is offered in the hope that honest modelling, shared publicly, contributes to better decisions being made before they are forced by crisis.

Josh Bardsley is the founder of FBED Labs. He builds intelligence systems that surface hidden patterns at scale. He lives in London with his daughter.