APPLIED ANALYSIS SERIES — AAS-003

THE DIRECTIVE™ — APPLIED ANALYSIS SERIES — AAS-003

AI in the Courts: Why Speed Without Safeguards Is Not Justice

A SAFECHAIN™ Governance Analysis of Artificial Intelligence, Court Backlogs, Disclosure Integrity, and Procedural Fairness

Reference: SAFECHAIN/AAS/2026/003

Author: Samantha Avril-Andreassen FRSA

Organisation: SAFECHAINN Ltd (Company No. 12038453)

Abstract

Artificial intelligence is increasingly proposed as a means of addressing court delays, administrative burdens, and growing case backlogs. The potential benefits are substantial. However, the central question is not whether artificial intelligence can increase efficiency. The question is whether efficiency alone constitutes reform.

This paper argues that speed without safeguards is not reform. Using selected papers from the SAFECHAIN™ Foundational Architecture Index™, it examines the governance questions raised by AI deployment within justice systems, in relation to disclosure, participation, historical data, accountability, and inter-agency coordination.

Keywords: Artificial Intelligence, Family Justice, Court Reform, Disclosure Integrity™, Participation Integrity™, Governance, Accountability, Coordination, SAFECHAIN™, The Directive™

1. Introduction: The Promise and Risk of AI in Justice

Across many jurisdictions, justice systems face the same challenge. Cases are taking longer. Court backlogs continue to grow. Administrative burdens consume increasing amounts of judicial and professional time. Governments therefore look toward technology as a means of improving efficiency.

Artificial intelligence is frequently presented as the next logical step. Large language models can review documents. Machine learning systems can identify patterns. Automated tools can sort evidence, organise case files, and generate administrative outputs at a speed no human team can match.

Viewed purely through the lens of productivity, the attraction is obvious. A process that currently takes days may be completed in minutes. A disclosure review involving thousands of documents may be filtered automatically. Routine administrative work may be delegated to software.

These developments have genuine potential. Yet justice systems are not manufacturing systems. The objective is not simply throughput. The objective is fair, lawful, and accountable decision-making. This distinction matters because courts exist to determine matters affecting liberty, safety, housing, finances, family relationships, and fundamental rights. An efficiency tool introduced into such a system cannot be evaluated solely on the basis of speed. It must also be evaluated on the basis of fairness.

2. The Backlog Argument

The case for artificial intelligence is often framed through backlog reduction. This argument has considerable force. Delays can be harmful. In family proceedings, delay can prolong uncertainty for children and families. In criminal proceedings, delay can affect victims, defendants, and witnesses. In civil proceedings, delay can increase costs and limit access to justice. Technology therefore appears attractive because it promises to reduce friction within the system.

The challenge, however, is that backlog reduction and justice are not identical objectives. A process can become faster while simultaneously becoming less reliable. A system can increase efficiency while decreasing transparency. A court can dispose of more cases while increasing the risk that vulnerabilities, safeguarding concerns, or evidential gaps are overlooked.

The question is therefore not "Can AI make courts faster?" The question is "Can AI make courts faster without compromising fairness, accountability, and participation?"

3. The Safeguarding Question

Every technological reform eventually reaches the same point: what happens when something important is missed? In many areas of government, missing a detail creates inconvenience. Within justice systems, missing a detail can create harm — a domestic abuse disclosure overlooked during document review, a safeguarding concern buried within records, a pattern of coercive control obscured by fragmented evidence, or a vulnerable individual's participation difficulties hidden within administrative data.

These are not hypothetical concerns. AAS-001 of this series demonstrated, with reference to Everyday Business (Domestic Abuse Commissioner, 2025), how a single structural gap — the C1A form's absence of a coercive control field — contributed to a fall from 81% of safeguarding letters noting domestic abuse allegations to only 64% validating them as relevant. That gap exists in a paper-based, human-mediated process. The question this paper examines is what happens to gaps of that kind when the filtering and validation steps are partly or wholly automated: do they become easier to detect and close, or harder?

Artificial intelligence cannot simply be assessed by asking whether it identifies most relevant information. It must be assessed by asking what happens when it misses the information that matters most. This is fundamentally a safeguarding question, and safeguarding questions require governance answers.

4. Disclosure Integrity™ and Automated Evidence Review

Paper 9 of the SAFECHAIN™ Foundational Architecture Index™ introduces the concept of Disclosure Integrity™. The core question is straightforward: can decision-makers rely upon the information that reaches them?

This question becomes significantly more important when artificial intelligence enters the disclosure process. AI systems are frequently proposed as tools for document filtering, evidence classification, relevance ranking, disclosure review, and summarisation. Each of these functions involves judgment. Even where the system is not making the final legal decision, it influences which information receives attention and which does not.

This introduces a critical governance challenge. An AI system that filters 100,000 documents into 1,000 documents has effectively shaped the evidential environment within which subsequent decisions are made. If relevant information is excluded during that process, the decision-maker may never know it existed. This does not require bad faith. It does not require technical failure. It simply reflects the reality that every filtering mechanism changes what remains visible.

Disclosure Integrity™ therefore requires questions that extend beyond technical performance:

• What information was excluded?

• Why was it excluded?

• Can exclusions be audited?

• Can decisions be challenged?

• Is there a complete record of what was filtered and why?

Without these safeguards, AI-assisted disclosure risks creating a new form of evidential opacity. The system may become faster. Yet it may simultaneously become more difficult to scrutinise. That is precisely the type of trade-off governance frameworks are designed to identify before implementation rather than after harm occurs.

5. Participation Integrity™ and Vulnerable Court Users

Paper 1 of the SAFECHAIN™ Foundational Architecture Index™ introduces The Participation Gap™ and the broader concept of Participation Integrity™. The central question is not whether a person is physically present within proceedings. The question is whether they are capable of participating effectively.

This distinction becomes increasingly important when technology is introduced into justice systems. Artificial intelligence is often described as neutral. Yet participation is rarely neutral. A legally represented commercial organisation may interact with digital systems very differently from a litigant in person. A professional institution may navigate automated processes very differently from a domestic abuse survivor experiencing trauma. A technologically confident individual may respond differently from someone with learning difficulties, language barriers, mental health challenges, or limited digital literacy. These differences already exist. The introduction of AI may either reduce them or amplify them. The outcome depends upon governance.

If AI-assisted systems generate summaries, recommendations, or procedural guidance, how can a vulnerable person understand how those outputs were reached? If automated systems identify procedural priorities, how can a participant challenge an incorrect classification? If vulnerability indicators are overlooked during automated review, how is that error identified and corrected?

Participation Integrity™ therefore requires more than accessibility. It requires meaningful challenge. It requires transparency. It requires procedural safeguards that ensure technological efficiency does not become a barrier to effective participation. A court process that is technically efficient but practically inaccessible to vulnerable participants cannot properly be described as fair.

6. Historical Data and the Passport of Erasure™

One of the most frequently discussed concerns surrounding artificial intelligence is bias. The issue is often misunderstood. Artificial intelligence does not develop bias in the same way humans do. Instead, AI systems learn patterns from data. This creates a different challenge.

If historical systems contain unequal outcomes, incomplete records, inconsistent safeguarding practices, or structural distortions, those patterns may be reflected within the data used to train or evaluate technological systems. The risk is not necessarily that AI becomes biased. The risk is that existing patterns become normalised. An algorithm cannot distinguish between what historically occurred and what should have occurred. It simply identifies patterns.

Paper 2 of the SAFECHAIN™ Foundational Architecture Index™, The Passport of Erasure™, examines how documentation loss, fragmentation, and institutional discontinuity erode an individual's identity, credibility, and access to systems over time. That paper was developed to describe what happens to a person as they move between institutions whose records of them are incomplete or inconsistent. The same structure applies, at a system level, to historical data used to train or calibrate AI tools: where the historical record itself reflects under-recognition of coercive control, minimisation of non-physical abuse, or inconsistent safeguarding practice, an AI system trained on that record inherits the erasure, and may reproduce it at scale and at speed.

This is an extension of Paper 2's reasoning from the individual case file to the training dataset, rather than a direct restatement of it, and is noted here as such. Governance of AI in justice systems should treat historical pattern replication as a Passport of Erasure™ question: not simply "is this dataset representative," but "whose absence from, or mischaracterisation within, the historical record does this dataset preserve?"

Practical governance questions include: what data was used? What outcomes does the system optimise? What assumptions are embedded within the model? How frequently is performance reviewed? What independent oversight exists? The objective should not be to eliminate technology. The objective should be to ensure that technology remains subject to the same scrutiny expected of any other decision-support mechanism within justice systems — particularly where domestic abuse, safeguarding, vulnerability, housing insecurity, financial abuse, or coercive control may be involved. Historical under-recognition of such issues should not become embedded within future technological systems.

7. Accountability: Who Is Responsible When AI Gets It Wrong?

Perhaps the most important governance question is accountability. When a human professional makes a mistake, responsibility can usually be identified. Processes exist for review. Reasons can be requested. Decisions can be challenged.

Artificial intelligence complicates this picture. If a relevant document is overlooked by an AI-assisted disclosure tool, who is accountable? The software provider? The court? The legal representative? The public authority? The individual operating the system?

These questions become even more difficult when AI outputs influence subsequent decisions without formally determining them. A recommendation may not be binding. A ranking system may not make the final decision. Yet both can shape human judgment.

This creates what SAFECHAIN™ Paper 22, The Accountability Paradox™, describes: the more complex systems become, the harder responsibility can become to identify. Paper 33, The Responsibility Paradox™, raises a related concern — when multiple actors contribute to an outcome, each may possess partial responsibility while no single actor accepts full responsibility. Artificial intelligence risks intensifying this dynamic.

Accountability must therefore be designed into systems before deployment. Not after problems emerge. Not after harm occurs. Not after public confidence has been damaged. Before implementation begins.

8. SAFECHAIN™ Governance Requirements for Court AI

The purpose of governance is not to prevent innovation. The purpose of governance is to ensure innovation remains accountable. Based upon the SAFECHAIN™ architecture, the following principles emerge.

Principle 1 — Disclosure Integrity™ (Paper 9)

All AI-assisted filtering, ranking, or review processes should be auditable. Users must be able to understand what information was included, excluded, prioritised, or deprioritised.

Principle 2 — Participation Integrity™ (Paper 1)

Participants must retain meaningful opportunities to challenge outputs that affect them. Technological efficiency must not replace procedural fairness.

Principle 3 — Continuity (Paper 26 — The Continuity Deficit™)

Safeguarding information must remain visible as a case moves between agencies, systems, and any AI-assisted stages within them. Technology should strengthen continuity rather than fragment it further.

Principle 4 — Accountability (Papers 22 and 33)

Human accountability must remain identifiable. Responsibility cannot be delegated to software.

Principle 5 — Coordination (Paper 25 — The Coordination Deficit™)

Where different courts, agencies, or tiers of the justice system adopt different AI tools, on different timetables, trained on different data, a new coordination risk arises in addition to the ones these tools are intended to solve. A case that moves between a court using one AI-assisted disclosure tool and an agency using another, or none, may experience a new form of discontinuity — not in the information itself, but in how that information is filtered, prioritised, or summarised at each stage. Coordination of AI deployment across the justice system is therefore not a secondary implementation detail; it is itself a governance question, on the same footing as the design of any individual tool.

Cross-cutting: Transparency

Transparency is not treated here as a sixth, freestanding principle alongside Principles 1–5, but as the condition that makes each of them meaningful in practice. Disclosure Integrity™ requires transparency about what was filtered; Participation Integrity™ requires transparency about how outputs were reached; Continuity requires transparency about what information moved, or did not move, between stages; Accountability requires transparency about who decided what. Affected individuals must understand how systems influence decisions that affect their rights, safety, liberty, finances, or family relationships — not as an additional requirement, but as the shared precondition for the other five.

These principles do not require opposition to artificial intelligence. They require responsible implementation.

9. Conclusion: Reform Must Be Measured by Fairness, Not Speed

Artificial intelligence will likely become a permanent feature of modern justice systems. The question is not only whether technology will be used, but how it will be governed.

The temptation within reform programmes is often to measure success through efficiency: cases processed, documents reviewed, backlogs reduced, costs lowered. These measures matter. But they are not the only measures that matter. Justice systems exist to deliver fair outcomes through fair processes. Any technological reform that increases efficiency while diminishing accountability, participation, safeguarding, transparency, or evidential reliability cannot properly be described as reform.

The challenge for policymakers considering AI-assisted tools in disclosure, case management, and administration is therefore not simply technological. It concerns how public institutions exercise power, how decisions are made, and how individuals are protected when those decisions affect their lives.

Artificial intelligence may assist justice. It may improve efficiency. It may reduce delay. Yet speed alone is not justice. Speed without safeguards is not reform.

Reading This Alongside the Architecture

This paper forms part of The Directive™ Applied Analysis Series and should be read alongside:

• Paper 1 — The Participation Gap™

• Paper 2 — The Passport of Erasure™ (extended, Section 6)

• Paper 9 — Disclosure Integrity™

• Paper 22 — The Accountability Paradox™

• Paper 25 — The Coordination Deficit™

• Paper 26 — The Continuity Deficit™

• Paper 33 — The Responsibility Paradox™

This paper should also be read alongside AAS-001 (Two Reports, One Chain), which provides the evidential basis for Section 3's account of disclosure and validation gaps in current, non-automated processes.

SAFECHAIN™ welcomes discussion with researchers, policymakers, technologists, legal professionals, regulators, and public institutions interested in exploring governance approaches that may complement technological innovation within justice systems.

© 2026 Samantha Avril-Andreassen. All rights reserved. SAFECHAINN Ltd (Company No. 12038453).

Version 1.0

Reference: SAFECHAIN/AAS/2026/003

Copyright & Intellectual Property Notice

© 2026 Samantha Avril-Andreassen. All rights reserved.

SAFECHAIN™, SAFECHAINN Ltd, The Directive™, Participation Integrity™, Passport of Erasure™, Shadow Ledger™, Coercive Debt Lifecycle™, Legacy Harm Architecture™, Institutional Failure Taxonomy™, Vulnerability Index™, Safeguarding Intelligence Model™, Seal of Integrity™, MØPIT™, SIP™, CPIT™, REBUILD™, COMPASS™, and all associated frameworks, methodologies, models, diagrams, terminology, research architecture, governance structures, assessment tools, training systems, and implementation mechanisms are proprietary intellectual property authored by Samantha Avril-Andreassen.

No part of this publication may be reproduced, copied, adapted, distributed, implemented, commercialised, taught, incorporated into training programmes, accreditation schemes, policy frameworks, software systems, artificial intelligence models, governance products, consultancy services, or derivative works without the prior written permission of the author.

The existence of this publication does not grant any licence to implement, replicate, modify, commercialise, or operationalise any SAFECHAIN™ intellectual property.

All rights reserved.