Leading the Charge: Why AI Literacy Must Be Mandatory in Criminal Law

A few years back, the notion that an algorithm might filter what counts as evidence in a criminal trial would have elicited polite scepticism, perhaps edged with some alarm. Fast-forward to now, and across the UK, this scenario is quietly unfolding. If you’re not fluent in AI yet, chances are you’re already lagging.

The legal system isn’t changing overnight, but it’s shifting faster than many professionals assume. At the heart of that shift is disclosure, which has long been the cornerstone of a fair trial.

The Disclosure Breakdown:  Where the System Buckles

The procedural fault lines of disclosure have long challenged criminal justice, cases delayed, prosecutions collapsing and trust eroding, whether among defendants or complainants.

In areas such as complex fraud, regulatory offenses, and major criminal cases, the volume of data has increased exponentially. Think millions of potential evidence items: emails, encrypted chats, calls, financial ledgers, images, the digital trail never ends.

Consider the Serious Fraud Office (SFO): one high-profile investigation generated a staggering 48 million documents, absorbing more than a quarter of its entire yearly budget.

This is where Technology‑Assisted Review (TAR) enters the picture.  AI built to sift, rank, and flag relevant documents for disclosure

TAR:  What It Is and Why It Matters

Predictive coding, or TAR, applies machine learning to classify documents.

Here’s how it works: first, a human labels a handful of documents. The AI learns from that and scales those judgments to vast datasets, improving over time through active feedback.

Typical techniques include:

• Recognising entities like names, dates, or organisations

• Clustering similar documents

• Expanding keywords to include synonyms or related concepts

While Courts haven’t formally mandated TAR, UK guidance, like Practice Direction PD57AD, originally for civil cases, now allows its use in complex fraud proceedings. The SFO has already run live pilots and plans broader deployment. But early missteps are already revealing the risks

The £3.3 Million Misfire

When the SFO introduced OpenText™ Axcelerate™, it was hailed as a breakthrough. Yet investigations and internal reviews uncovered something alarming: the tool tripped over punctuation. If it saw “bribe.” (with a full stop), it might entirely skip that document.

That oversight could mean crucial evidence, sometimes even exonerating information, never sees human eyes. Over 60 cases were impacted, denting confidence and reinforcing fears that hidden AI flaws may precipitate untraceable injustices. This wasn’t the tool’s first stumble; it also derailed the SFO’s high-stakes G4S prosecution in 2019.

On the Frontlines: Flaws, Fatigue, Fallout

The SFO isn’t alone in feeling strain.

The Crown Prosecution Service and police forces are overwhelmed:

• Disclosure teams often receive less than a week’s legal training

• Workloads pile up, caseloads multiply

• Non-lawyers frequently handle decisions that demand legal acumen

A 2025 Government review exposed some brutal stats:

• 306 officer-years wasted annually on cases that fell apart

• Over 210,000 hours spent redacting materials for dropped or dismissed cases

High-profile trials have unravelled amid failures to hand over key evidence, leaving victims with no answers and defendants tethered to long waits.

Can AI Be the Answer?

What AI does well:

• It can plough through mountains of data in minutes, not months.

• It doesn’t get tired, distracted, or biased the way humans sometimes are.

• It can surface patterns we might otherwise miss.

Where it falls short:

• AI lacks subtlety; it doesn’t interpret motive, tone, or nuance like a seasoned lawyer can.  That said, at the rate of progress, it won’t be long before it can.

• TAR is a “black box”: you get results, but not the reasoning. If something gets missed, you can’t trace how or why.

• Judges and defence lawyers, often unfamiliar with TAR, aren’t equipped to probe its methods.

In short, AI can deliver efficiency, but only transparency can uphold justice.

Cold Cases and a New Ethical Faultline

AI’s not just helping with disclosure, it’s being used to crack cold cases. The National Police Chiefs’ Council tested a system called Söze, which tackled 27 cold cases in under 30 hours, work that would have taken human teams roughly 81 years to complete.

Impressive? Certainly. But only if the training data and assumptions are solid. When they aren’t, the consequences are grave:

• False positives could ruin innocent lives

• False negatives might let offenders roam free

The New Divide: AI‑Fluent vs. Left Behind

Here’s the uncomfortable truth:

“Arriving at a digital crime scene armed with only a notepad and biro.”

If your legal defence team doesn’t understand tools like TAR, they cannot question them or guard against mistakes. They can’t protect clients or ensure fairness.

And the gap is growing. AI integration into prosecutor workflows is accelerating, likely full-scale within 18 months. A defence that remains analogue isn’t just behind; it’s systemically disadvantaged.

Are We Prepared?

Some institutions are making moves:

• The Law Society’s 2024 AI Strategy launched frameworks for innovation, integrity, and responsible adoption, plus AI literacy guides.

• The Bar Council rolled out use‑of‑AI guidance, formed an AI working group, and even banned AI-generated submissions for pupillages.

Still, significant gaps remain:

• No mandatory CPD training in AI for criminal lawyers

• No funded access to TAR tools for the defence

• Rules of procedure haven’t caught up with oversight of AI tools in the Court

If prosecutor systems are becoming digitally enhanced, the defence must be too, or risk being silenced.

What Must Change: Right Now

To preserve fair trials, we need to:

1. Demand explainable AI systems that leave reasoning and logic open to audit

2. Build AI fluency across judges and lawyers, make it a basic competence

3. Ensure the defence has access, through funding, tools, or partnerships

4. Update legislation, bring the Criminal Procedure Rules into the digital age, with oversight for algorithmic tools

The Volume Illusion: How Legal Aid’s Economic Model Is Set to Collapse

Beneath the surface of the criminal legal aid system lies a financial balancing act that most outsiders rarely see. It's not just that some cases pay poorly; it’s that most cases do, especially at the lower end of the criminal spectrum. Minor offences, police station attendances, or Magistrates’ Court hearings are often run at a loss or for minimal profit under fixed fee regimes. And yet, practitioners continue to take them on. Why? Because the system has, until now, relied on a swings-and-roundabouts model: the occasional serious case with a high page count offsets the unsustainable economics of the rest.

Under the Litigators’ Graduated Fee Scheme (LGFS), remuneration scales with volume. The more pages served, the higher the fee. VHCCs, sprawling, document-heavy prosecutions involving fraud, drug conspiracies, or encrypted communications, have propped up many defence practices, not because they’re efficient, but because they’re financially dense. A 10,000-page case might be a logistical headache, but at least it pays.

But here’s the break point: AI collapses that model. If an algorithm can triage and extract relevance from 10,000 pages in minutes, what happens to the economic assumptions baked into current funding? In VHCCs, legal aid firms agree hours in advance with the LAA based on estimated human review time. Consider: 10,000 pages of mobile phone data at two minutes per page equals 333 hours of billable time. If AI can process it in under an hour, with a few hours of human oversight, the solicitor is no longer rewarded for rigour but penalised for efficiency.

In this context, AI doesn’t improve margins; it destroys them. The more advanced your workflow, the faster your analysis, the less your case is worth under legacy legal aid structures. You get paid less for doing better work and fast becomes synonymous with undervalued.

This is not a theoretical risk. As AI becomes embedded in prosecutorial tools, the defence is being left behind, not for lack of will, but because the funding model punishes innovation. The very architecture of legal aid is antithetical to modernisation. And without urgent reform, the principle of equality of arms begins to erode, not in the abstract, but in real trials, real lives.

What emerges is a growing chasm between those with access to innovation and those without. Private defence firms, or mixed practices able to invest in tech, train their teams, and adapt at speed, are already pulling away. Legal aid firms, bound by capped contracts and frozen fee structures, are stuck on analogue rails. Some are quietly pivoting toward private work, not by strategy but by necessity, because survival now requires stepping outside the system that once sustained them.

If this continues, we’ll have two tiers: one digitally empowered, the other digitally excluded. And in that world, the quality of your representation won’t be about justice; it’ll be about whether you can afford it.

AI Isn’t the Enemy, But Ignorance, Inequality and Incentive Failure Are

AI is going to transform how criminal law is practiced. The question is no longer whether we will use it, but whether we will use it wisely, transparently and equitably.

At one end, we face a growing skills divide. If defence lawyers don’t understand how algorithmic systems work, they can’t challenge them. If judges don’t know what TAR is doing under the hood, they can’t ensure fairness. And if prosecutors are armed with digital tools while defence teams remain without, we don’t just lose efficiency, we lose balance.

But there’s another divide, quieter, but just as dangerous: the economic disincentive to innovate. Legal aid still pays for page-turning, not insight. The most financially sustainable cases today are the ones that AI can reduce to minutes of work. Without reform, the firms that dare to modernise risk eroding their own revenue on the VHCC cases and the assumptions upon which the fixed fee cases are based, no longer apply. Progress, perversely, may come at a price they can’t afford to pay.

If we don’t fix this, from funding models to training mandates, we will build a two-tier system: one where the prosecution uses smart tools to prosecute faster and cheaper, and the defence is expected to resist it with biro and a binder.  Unless of course, a defendant has the means to instruct a top firm on a private basis.

In digital justice, if you’re not ahead of the curve, you’re already behind. But without structural change, many won’t even have the chance to compete.

How Forbes Solicitors Can Help

As AI takes root in the criminal justice system, from digital disclosure tools to algorithm-assisted investigations, the legal landscape is undergoing a quiet revolution. And with that change comes risk: ethical, procedural, and reputational.

At Forbes Solicitors, we help clients navigate this complex new reality. Whether you’re defending a case where AI tools influenced disclosure decisions, challenging the reliability of automated evidence review, or preparing for litigation involving algorithmic missteps, we’re ready to support you.

We advise private clients, professionals, and high-stakes defendants on:

• Challenging flawed or biased AI disclosure processes

• Navigating cases impacted by TAR and predictive coding tools

• Preparing expert-led defence strategies in digital-first investigations

• Building AI literacy within legal teams to counter systemic disadvantage

• Mitigating risk under emerging rules around algorithmic evidence and digital fairness

Our High-Profile and Private Crime Division, led by Craig MacKenzie, brings a rare combination of criminal law expertise and forward-thinking insight. We work with discretion, speed, and strategic clarity to protect our clients in this rapidly evolving space.

Our firm also has one of the largest legal aid teams in the country, dealing with more than 5000 criminal cases per year.  We invest heavily in training and technology to remain ahead of the curve.

If you’re facing legal challenges involving algorithmic justice or want to proactively safeguard your rights in AI-driven prosecutions, contact Craig MacKenzie via email below or call 01772 220022.