In Response to the Ministry of Justice AI Pilot Announcement, 9 June 2026
THE DIRECTIVE™ · SAFECHAIN™ · JUNE 2026
Policy Analysis · In Response to the Ministry of Justice AI Pilot Announcement, 9 June 2026
Speed Without Safeguards
Is Not Reform.
Why the Ministry of Justice AI Pilot Exposes the Next Governance Challenge and Why the SAFECHAIN™ Algorithmic Accountability Standard Is Required
Author: Samantha Avril-Andreassen FRSA · LLB (Hons) · LLM · LPC
Founder & CEO, SAFE-CHAINN Ltd · Company No. 12038453 · samantha@safe-chain.org
This Directive responds to the Ministry of Justice announcement of 9 June 2026, made by the Deputy Prime Minister David Lammy at London Tech Week, of AI legal assistants and AI-assisted case listing tools to be trialled in Crown Court proceedings in England and Wales. It sets out the SAFECHAIN™ governance position: that the deployment of algorithmic tools within the justice system requires — before efficiency — accountability architecture, participation safeguarding standards, equality impact assessment, and the institutional capability to ensure that the speed gained does not come at the cost of the safeguarding conditions that make justice possible. It proposes the SAFECHAIN™ Algorithmic Accountability Standard — five governance criteria that must be satisfied before AI deployment in justice proceedings is extended beyond sandbox testing. Not legal advice.
© 2026 Samantha Avril-Andreassen FRSA. All rights reserved. All SAFECHAIN™ frameworks are protected under UK copyright and intellectual property law.
Section 1 — The Announcement
What the Ministry of Justice Has Proposed
On 9 June 2026, Deputy Prime Minister and Secretary of State for Justice David Lammy announced at London Tech Week that the Ministry of Justice will trial AI legal assistants in Crown Courts in England and Wales. The tools, developed in partnership with UK legal experts and leading AI developers, are described as digital paralegals — intended to support legal professionals with routine casework including research and case analysis, the summarising of documents, the identification of cases ready for trial, and the grouping of similar hearings to maximise judicial and prosecutorial resource.
The announcement was made in the context of a Crown Court backlog that has reached a record 80,000-plus cases, with some trials not currently listed until 2030. The MoJ has framed the AI pilot as one component of a broader Justice AI Action Plan, the stated objectives of which include strengthening AI foundations, embedding AI across services through a Scan, Pilot, Scale model, and — by 2027 — delivering system-wide AI integration at scale. The stated guiding principle of the Action Plan is to put safety and fairness first. The tension between that stated principle and the pace of the proposed deployment is the governance question this paper addresses.
The AI pilot is not without precedent within the MoJ's own operations. Justice Transcribe — an AI tool that records and transcribes offender meetings — has already been deployed across the probation service, with the department reporting projected savings equivalent to 18,750 days of staff time annually. A similar tool is being trialled in immigration and asylum tribunals. Free sentencing-remark transcripts for Crown Court victims are promised from spring 2027.
The Ministry of Justice is not proposing to introduce AI into justice proceedings. It has already introduced AI into justice proceedings. The Crown Court pilot extends a programme that is already operational. That is precisely why the governance question is urgent.
The Law Society, representing more than 200,000 solicitors, has welcomed the pilot but stated that AI cannot replace vital funding and additional court staff, and that pilot evaluations must be thorough and publicly reported. This paper endorses both of those positions and proposes the specific governance architecture that would make them operational rather than aspirational.
THE ANNOUNCEMENT — KEY DETAILS
Date: 9 June 2026 · Venue: London Tech Week · Announced by: David Lammy, Deputy Prime Minister and Secretary of State for Justice
Functions: AI legal assistants for routine research and case analysis; AI case listing tool to identify trial-ready cases and group similar hearings; Justice Transcribe already operational in probation; trial in immigration and asylum tribunals ongoing
Backlog context: 80,000+ Crown Court cases; some trials not listed until 2030
Testing model: AI Growth Labs — sandbox environments for safe and controlled testing before wider rollout
MoJ position: AI will play no role in judicial decision-making; tools will meet standards required by judges and lawyers before rollout
Law Society: AI cannot replace vital funding and court staff; evaluations must be thorough and publicly reported
Section 2 — The Documented Failures
What Has Already Gone Wrong
The MoJ announcement was made in the same week that the legal sector continues to process documented instances of AI-generated errors that have caused direct harm to legal proceedings. These are not theoretical risks. They are live failures, already on the record, in UK and comparable jurisdictions. Their significance for the governance argument is fundamental: they demonstrate that the question of AI deployment in justice is not a future governance challenge. It has already produced harm. The governance architecture required to prevent further harm must be established before, not after, deployment at scale.
The Hallucination Problem
In an £89 million case against Qatar National Bank in the English courts, 18 of 45 legal authorities cited in submissions were fictitious. They had been generated by a public AI tool and presented as genuine case law. The court's ability to adjudicate was directly compromised by the AI system's confident production of plausible-sounding but non-existent legal authority.
In a Haringey housing case, phantom case law was cited five times in submissions. Again, generated by AI, presented as genuine, and unchallenged until identified by the court. The parties — at least one of whom was likely a litigant in person without the legal expertise to identify fabricated authority — had their case heard on the basis of a false legal landscape.
A Microsoft Copilot hallucination — the generation of a non-existent match — helped justify a football policing decision. The episode was sufficiently serious that it prompted guidance from the relevant authorities pausing the use of AI in police court statements. The guidance exists. The deployment continues.
These three incidents share a structural characteristic that is directly relevant to the MoJ pilot. In each case, the AI system did not flag uncertainty. It did not identify its output as speculative or unverified. It produced confident, plausible, professional-sounding output that was materially false. The capacity of AI systems to hallucinate with confidence — to generate fiction that is indistinguishable in form from accurate information — is not a technical glitch awaiting resolution. It is a documented feature of the technology being deployed.
An AI system that fabricates case law is not a procedural inconvenience. It is a structural threat to the administration of justice. The question is not whether the MoJ has adequate controls. The question is whether those controls are sufficient when deployed against a technology that fails with confidence rather than flagging its own uncertainty.
The Equality of Arms Problem
The MoJ pilot proposes AI tools developed in partnership with legal experts. Those tools will, in the first instance, be available to legal professionals — prosecutors, defence solicitors, court staff. They will not, by definition, be available to litigants in person who represent themselves in Crown Court proceedings, family court proceedings, or tribunal hearings without access to legal representation.
The Crown Court backlog of 80,000 cases contains a substantial proportion of unrepresented parties. The removal of legal aid from the majority of private family law proceedings under the Legal Aid, Sentencing and Punishment of Offenders Act 2012 created a cohort of litigants in person who lack the professional expertise to identify AI-generated hallucinations, to challenge AI-assisted submissions, or to access the same AI tools that the represented party's legal team is using.
Where AI tools enhance the capability of the represented party without providing equivalent capability to the unrepresented party, the deployment of AI does not narrow the equality of arms gap. It widens it. A technology announced as reducing the backlog and improving access to justice, deployed in a manner that materially increases the informational and analytical advantage of represented parties over unrepresented ones, produces the opposite of its stated objective for the people it most needs to serve.
Section 3 — The Legal Framework
The Obligations That Already Apply
The MoJ's stated guiding principle — put safety and fairness first — is not merely aspirational. It is a description of obligations that are already legally binding on the Ministry of Justice, on HMCTS, and on the courts as public authorities. The deployment of algorithmic tools in justice proceedings does not create new legal obligations. It engages obligations that already exist. The governance question is whether the current pilot architecture is designed to discharge them.
Human Rights Act 1998 — Article 6
Article 6 of the European Convention on Human Rights, given domestic effect by the Human Rights Act 1998, guarantees the right to a fair hearing. The European Court of Human Rights has consistently interpreted this as requiring equality of arms — each party must have a reasonable opportunity to present their case in conditions that do not place them at a substantial disadvantage vis-a-vis the opposing party. Where AI tools are deployed asymmetrically — available to one party or to court staff in ways that materially affect the information and analysis available to the proceedings — the equality of arms requirement is directly engaged.
Section 6 of the Human Rights Act 1998 makes it unlawful for a public authority to act in a way incompatible with a Convention right. Courts are public authorities. HMCTS is a public authority. The MoJ is a public authority. The obligation to ensure that AI deployment does not create or exacerbate conditions incompatible with the right to a fair hearing is not a governance aspiration. It is a legal obligation already binding on every institution involved in the pilot.
Equality Act 2010 — Section 149
The Public Sector Equality Duty under section 149 of the Equality Act 2010 requires public authorities to have due regard to the need to advance equality of opportunity and eliminate discrimination. The deployment of AI tools in Crown Court proceedings requires — as a matter of statutory obligation, not discretionary good practice — a formal equality impact assessment. This assessment must consider the differential impact of AI deployment on parties with protected characteristics, including those whose protected characteristics are directly associated with the domestic abuse and safeguarding contexts this paper addresses.
Where AI tools are trained on historical case data that reflects existing judicial bias — the kind of bias documented by the Right to Equality report published the same day as the MoJ announcement — those tools may systematically reproduce and entrench that bias at algorithmic speed. An AI system trained on judgments that contain victim-blaming language and that minimise coercive control does not correct for those patterns. It learns from them. The s.149 duty requires that this risk be assessed and mitigated before deployment, not identified through retrospective audit.
Data Protection Act 2018 and UK GDPR — Article 22
Article 22 of the UK GDPR provides that data subjects have the right not to be subject to a decision based solely on automated processing where that decision produces legal effects or significantly affects them. The MoJ's position that AI will play no role in judicial decision-making is an important commitment. But the line between AI-assisted analysis that informs a judicial decision and automated processing that produces legal effects is not as clear as the current framing suggests. Where AI tools summarise evidence, identify cases as trial-ready, or group similar hearings, those outputs directly affect the procedural trajectory of individual cases. The Art.22 framework requires that data subjects whose cases are affected by automated processing be informed, and that appropriate safeguards be in place.
Victims and Courts Act 2026
The Victims and Courts Act 2026, strengthening accountability mechanisms for the treatment of victims within the justice system, creates new obligations that are directly relevant to AI deployment. The Act's provisions on victims' rights and transparency cannot be discharged by a system that processes cases faster if that system is incapable of recognising the safeguarding context within which the case arises. A victim of coercive financial control whose case is processed by an AI system trained on historical data does not benefit from the Act's provisions if the AI system that assists with their case is incapable of identifying economic abuse, participation impairment, or the Information Control Doctrine patterns that characterise their circumstances.
The Case Law Framework
Case
Principle and Application
R (Bridges) v Chief Constable of South Wales Police [2020] EWCA Civ 1058
The Court of Appeal held that the use of automated facial recognition technology by police without an adequate legal framework, a data protection impact assessment, or proper equality impact assessment was unlawful. The court confirmed that algorithmic tools used in justice contexts must satisfy proportionality, necessity, and equality obligations under existing law. The principles apply directly to AI deployment in Crown Court proceedings.
Uber BV v Aslam [2021] UKSC 5
The Supreme Court on algorithmic management and accountability: where algorithmic systems are used to make or substantially influence decisions affecting individuals, those individuals are entitled to transparency, challenge, and accountability. The principle that algorithmic systems do not displace the accountability obligations of the institutions deploying them is directly applicable to justice AI.
Big Brother Watch v United Kingdom [2021] ECHR
The European Court of Human Rights on automated surveillance systems and Art.6/Art.8: the deployment of automated tools by public authorities must be subject to adequate legal basis, proportionality assessment, and effective judicial oversight. Confirms that the deployment of technology by justice institutions does not create a carve-out from Convention obligations.
Qatar National Bank hallucination case [2024-2026]
Not a precedent in the formal sense, but a documented example with direct evidential significance: AI-generated hallucinations in legal submissions — 18 of 45 cited authorities fictitious — demonstrate that the risk is not theoretical. Courts and legal systems deploying AI must have governance mechanisms capable of identifying and correcting AI-generated error before it affects judicial decisions.
Section 4 — The SAFECHAIN™ Analysis
What AI Cannot See — and Why That Is a Governance Problem
SAFECHAIN™ is not anti-technology. The position of this paper is not that AI has no place in the justice system. It is that AI deployment in justice proceedings requires a governance architecture capable of ensuring that the speed gained does not come at the cost of the safeguarding conditions that make justice possible. That governance architecture does not currently exist. This section identifies, through the SAFECHAIN™ framework, the specific dimensions along which the current pilot is insufficient.
The Knowledge-to-Harm Pathway™ Applied to Algorithmic Speed
The SAFECHAIN™ Knowledge-to-Harm Pathway™ maps the five-stage sequence through which institutional knowledge fails to become institutional protection: Knowledge, Foreseeability, Capacity, Inaction, Harm. Applied to the deployment of AI in justice proceedings, the Pathway operates with a specific and concerning modification: at algorithmic speed, the interval between the stages collapses.
A human legal professional who misses a safeguarding indicator in a case file can, through the next hearing, the next review, the next case management conference, be brought back to the evidence and given the opportunity to correct. The correction mechanism is built into the procedural architecture of the justice system because the system was designed around human fallibility and the time required to recognise and address it.
An AI system that misses a safeguarding indicator does not have the same correction opportunity. It processes the case, produces its output, identifies the case as ready for trial or suitable for grouping, and moves to the next one. The speed that is the objective of the deployment is also the mechanism by which the correction window closes. The Knowledge-to-Harm Pathway™, in an AI-accelerated environment, produces harm at algorithmic speed — and accountability mechanisms designed for human-paced proceedings cannot keep up.
The Participation Integrity™ Problem
The SAFECHAIN™ Participation Capacity Variability™ (PCV™) model identifies five factors that determine whether a party can participate effectively in proceedings: trauma response, cognitive load, financial resource, institutional familiarity, and documentation access. Each of these factors is, in whole or in significant part, invisible to a document-processing AI system.
Trauma does not appear in a disclosure schedule. It appears in the non-linear narrative of a victim's evidence, in the inconsistencies that arise from memory fragmentation under hypervigilance, in the reluctance to name specific events that a legal professional trained in trauma response would recognise as a signature of sustained abuse. An AI system trained to identify relevant documents, summarise evidence, and flag procedural readiness does not have, and cannot acquire from document analysis, the capability to identify that the party whose case it is processing is participating from a position of profound cognitive and emotional impairment.
Economic abuse does not appear in documents as economic abuse. It appears in the absence of documents — in the missing bank records, the absent pension statements, the company structures that do not appear in Form E because the Information Control Doctrine has operated to exclude them. An AI system that analyses what has been filed cannot identify what has not been filed. Its analysis of disclosure completeness is bounded by what it can see. The Shadow Ledger™ — the parallel financial reality maintained by the concealing party — is, by definition, invisible to a system that analyses the disclosed picture.
This is not a criticism of AI capability. It is a description of the specific mismatch between what AI can do and what justice in these cases requires. Document analysis, legal research, and procedural case management can be assisted by AI. The assessment of whether genuine participation is possible — the Participation Integrity™ standard — cannot.
The Information Control Doctrine™ Applied to AI-Assisted Proceedings
The SAFECHAIN™ Information Control Doctrine™ identifies four stages through which coercive control dismantles the informational conditions that fair proceedings require: information acquisition, consolidation, weaponisation, and litigation deployment. At Stage 4, the informational advantage constructed during the relationship is actively deployed within the proceedings themselves.
In an AI-assisted proceedings environment, Stage 4 of the Information Control Doctrine™ acquires a new and specific danger. The party whose legal team has access to AI tools for disclosure analysis, evidence summarising, and case preparation holds a Stage 4 information control advantage over the litigant in person who does not. The weaponisation of informational advantage — already documented as a feature of high-conflict proceedings — is accelerated, expanded, and deepened by differential access to AI capability.
The MoJ pilot does not address this. Its stated objective is to deploy AI tools developed with legal experts to support legal professionals with routine casework. Legal professionals already have the advantage of professional training, institutional access, and case experience. Adding AI capability to an already advantaged party, without simultaneously considering how the unrepresented party will be protected from the consequences of that widening advantage, is not reform. It is acceleration of the existing inequality.
Documentation Continuity™ and Safeguarding Context
The SAFECHAIN™ Documentation Continuity™ standard addresses the specific safeguarding risk created when evidential records pass through multiple institutional hands without the contextual information that makes them meaningful for safeguarding purposes. Applied to AI processing, the risk is specific and consequential.
An AI system that summarises a case file produces a summary. That summary is a reduction of the original material. Every reduction involves a selection — a decision, embedded in the algorithm, about what is material and what is not. Where that selection was made by a model trained on historical legal data, the selection reflects the priorities embedded in that data. In proceedings involving domestic abuse, coercive control, and economic abuse, those priorities may systematically underweight safeguarding context — the very context that the Right to Equality report, published the same day as the MoJ announcement, has documented is already being systematically missed by human judges.
An AI-assisted proceedings architecture that summarises cases faster, but whose summaries systematically omit or underweight the safeguarding context that human decision-makers are already failing to recognise, does not improve justice. It accelerates injustice. Documentation Continuity™ requires that AI processing preserve — not reduce — the safeguarding context within case material.
Section 5 — The Governance Standard
The SAFECHAIN™ Algorithmic Accountability Standard
The SAFECHAIN™ Algorithmic Accountability Standard proposes five governance criteria that must be positively satisfied before AI deployment in justice proceedings moves beyond sandbox testing and into operational use. These criteria are not aspirational. Each derives from an existing legal obligation — under the Human Rights Act, the Equality Act, the Data Protection Act, or the common law. The Standard does not propose new law. It proposes the operational governance architecture required to discharge existing law in the context of AI deployment.
Standard 01 — PARTICIPATION INTEGRITY ASSESSMENT
Before deployment: Can this system identify the conditions under which a party cannot participate effectively — including trauma response, economic abuse, PCV factors, and the participation barriers created by coercive control?
Current position: No participation integrity assessment is required or proposed under the current MoJ pilot framework.
Required: Before any AI tool is deployed beyond sandbox testing in proceedings involving domestic abuse, coercive control, or vulnerability indicators, the deploying authority must demonstrate that the tool has been assessed for its capacity to identify or preserve participation integrity signals — and must establish alternative safeguards for the conditions the tool cannot assess.
Standard 02 — EQUALITY IMPACT ASSESSMENT
Before deployment: Has a formal equality impact assessment been conducted, specifically addressing differential impact on protected characteristics, differential access to AI tools between represented and unrepresented parties, and the risk of bias reproduction from historical training data?
Current position: The MoJ's stated guiding principle is to put safety and fairness first. No formal published equality impact assessment of the Crown Court AI pilot has been identified.
Required: A formal equality impact assessment under EA 2010 s.149, specifically addressing algorithmic bias reproduction and the equality of arms consequences of asymmetric AI access, must be published before any Crown Court AI deployment moves beyond the sandbox stage.
Standard 03 — HALLUCINATION GOVERNANCE PROTOCOL
Before deployment: What specific mechanisms exist to identify, flag, and correct AI-generated errors — including hallucinated case law, fabricated authorities, and incorrect procedural analysis — before those errors affect judicial decisions?
Current position: The documented hallucination incidents — 18 fictitious authorities in the Qatar National Bank case; phantom case law five times in the Haringey housing case; AI hallucination in football policing — demonstrate the risk is live. Current MoJ sandbox testing does not have a published hallucination governance protocol.
Required: A published hallucination governance protocol, including verification requirements before AI-assisted outputs are used in submissions or judicial analysis, mandatory disclosure of AI assistance, and accountability mechanisms for AI-generated errors, must be in place before deployment.
Standard 04 — DOCUMENTATION CONTINUITY STANDARD
During deployment: Is AI processing designed to preserve — not reduce — safeguarding context within case material? Are summaries and analyses produced by AI tools required to flag safeguarding indicators, domestic abuse context, coercive control patterns, and vulnerability indicators present in the source material?
Current position: The current pilot proposes AI tools for summarising documents and case analysis. No safeguarding-context preservation standard has been published for those tools.
Required: AI tools deployed in proceedings involving domestic abuse must be required to operate to a Documentation Continuity standard: summaries and analyses must specifically preserve and flag safeguarding context, not reduce it through generic summarisation that reflects historical legal data priorities.
Standard 05 — ACCOUNTABILITY ARCHITECTURE
Throughout deployment: Can decisions made with AI assistance be audited, challenged, and — where AI error has caused harm — held to account? Do parties whose cases were processed by AI tools have the right to know, the right to see, and the right to challenge?
Current position: The MoJ has stated that AI will play no role in judicial decision-making. The line between AI-assisted analysis and judicial decision-making is procedurally important but does not resolve the accountability question: where AI-assisted case preparation has affected the trajectory of proceedings, the affected party must be able to identify, understand, and challenge that effect.
Required: A published accountability architecture for AI-assisted proceedings, including disclosure obligations, audit rights, and challenge mechanisms for AI-affected case handling, must be in place before Crown Court deployment.
Section 6 — The Efficiency Argument
Answering the Counter-Argument
The obvious counter-argument to the governance position set out in this paper is that the backlog is the most urgent justice problem facing the Crown Court, that 80,000 cases with trials listed until 2030 represents a systemic failure with direct human consequences, and that governance requirements for AI deployment — however well-founded — must not be allowed to prevent the innovation required to address that failure.
SAFECHAIN™ accepts the urgency of the backlog. It does not accept that the urgency of the backlog displaces the governance obligations that apply to the proposed solution. For two reasons.
First, the hallucination evidence. A case processed through an AI system that produces fictitious case law, summarises evidence in ways that miss safeguarding context, or generates analysis that reflects the biases documented in the Right to Equality report, is not a case whose backlog problem has been solved. It is a case whose errors will generate appeals, retrials, set-aside applications, and judicial review proceedings that increase the backlog rather than reducing it. The governance architecture proposed in this paper is not a barrier to efficiency. It is the precondition for efficiency that is durable rather than cosmetic.
Second, the equality argument. A Crown Court AI deployment that widens the equality of arms gap between represented and unrepresented parties is not solving the justice crisis facing the people most affected by the backlog. The 80,000 cases include a substantial proportion involving domestic abuse, coercive control, and vulnerable parties. If the AI tools deployed to process those cases accelerate the procedural machinery while increasing the informational advantage of the represented party, the cases move faster to outcomes that are less just. Faster injustice is not reform.
The question is not whether courts should use AI. The question is whether AI will be governed by safeguarding principles capable of protecting the people the justice system exists to serve. Governance is not the obstacle to reform. It is the foundation of reform that lasts.
Section 7 — Formal Submissions
What SAFECHAIN™ Is Asking the Ministry of Justice to Do
This paper makes formal governance submissions to the Ministry of Justice, HMCTS, and the Justice AI Unit. Each submission is grounded in the legal obligations identified in Section 3 and the governance analysis of Sections 4 and 5.
To the Ministry of Justice
1. Publish a formal equality impact assessment under EA 2010 s.149 for the Crown Court AI pilot before any deployment beyond sandbox testing, specifically addressing: algorithmic bias reproduction from historical training data; differential access to AI tools between represented and unrepresented parties; and the disproportionate impact of AI deployment on parties with protected characteristics in domestic abuse and coercive control contexts.
2. Commission an independent assessment of the Crown Court AI pilot against the SAFECHAIN™ Algorithmic Accountability Standard — five criteria — before any rollout beyond controlled testing environments.
3. Publish a Hallucination Governance Protocol establishing verification requirements, disclosure obligations, and accountability mechanisms for AI-generated errors in court proceedings.
4. Consider the equity of access implications of AI deployment: where AI tools are available to legal professionals but not to litigants in person, the MoJ should assess whether additional support — including AI-assisted tools or enhanced advocacy support — should be made available to unrepresented parties to maintain equality of arms under HRA Art.6.
To HMCTS and the Justice AI Unit
5. Adopt the SAFECHAIN™ Documentation Continuity™ standard as a requirement for AI tools deployed in proceedings involving domestic abuse indicators: AI summaries and analyses must specifically preserve and flag safeguarding context.
6. Establish a Participation Integrity Assessment requirement within the AI pilot evaluation framework, specifically addressing the capacity of AI tools to identify or preserve participation integrity signals in cases involving coercive control, economic abuse, and vulnerability.
7. Publish the outcomes of all pilot evaluations — including failure rates, hallucination incidents, equality impact data, and safeguarding signal preservation assessments — as a condition of any rollout beyond the sandbox stage.
8. Engage with the SAFECHAIN™ governance framework as a reference architecture for the accountability and safeguarding dimensions of the Justice AI Action Plan.
THE SAFECHAIN™ POSITION
The Ministry of Justice AI pilot represents genuine ambition to address a justice crisis. The Crown Court backlog is real. Its human consequences are severe. Technology may be part of the solution. But the governance architecture required to ensure that AI deployment produces just outcomes — not merely faster ones — does not yet exist within the pilot framework. The SAFECHAIN™ Algorithmic Accountability Standard provides five criteria that must be satisfied before deployment extends beyond controlled testing. They derive from existing legal obligations. They do not require new legislation. They require institutional will to apply obligations that are already binding — before the harm that will otherwise follow is irreversible. Faster injustice is still injustice. And safeguarding must remain infrastructure.
SAFECHAIN™ Framework Reference
Frameworks Deployed in This Paper
Framework
Application in This Paper
Knowledge-to-Harm Pathway™
Applied to algorithmic speed: the five-stage Pathway — Knowledge → Foreseeability → Capacity → Inaction → Harm — operates at algorithmic speed in AI-assisted proceedings, collapsing the correction window available in human-paced proceedings.
Participation Capacity Variability™ (PCV™)
Five participation barriers — trauma response, cognitive load, financial resource, institutional familiarity, documentation access — that AI document-processing tools cannot identify or assess. The analytical basis for Standard 01 of the Algorithmic Accountability Standard.
Participation Integrity™
The governance standard requiring positive assessment of whether genuine participation is possible before proceedings advance. Proposed as a required dimension of AI pilot evaluation frameworks.
Information Control Doctrine™
Stage 4 — Litigation Deployment — is amplified in AI-assisted proceedings where differential access to AI tools deepens the informational advantage of the represented party over the litigant in person.
Shadow Ledger™
An AI system that analyses disclosed material cannot identify the Shadow Ledger — the parallel financial reality maintained through non-disclosure. Directly relevant to the limits of AI-assisted disclosure analysis.
Documentation Continuity™
The standard requiring that AI processing preserve safeguarding context within case material. Standard 04 of the Algorithmic Accountability Standard. Addresses the specific risk that AI summarisation reduces or omits safeguarding signals present in source material.
SAFECHAIN™ Algorithmic Accountability Standard
Five-criteria governance framework: Participation Integrity Assessment, Equality Impact Assessment, Hallucination Governance Protocol, Documentation Continuity Standard, Accountability Architecture. The operational governance instrument proposed by this paper.
Institutional Inertia Paradox™
Applied to AI governance: the five conditions — knowledge without accountability, governance fatigue, siloed accountability, legitimacy through procedure, purpose drift — that produce institutional inaction in the face of documented AI risks.
Contact and Document Requests
– samantha@safe-chain.org (subject: AI in the Courts — SAFECHAIN Governance Response)
– safe-chain.org · safe-chain.org/pilot-application
SAFECHAIN™ · SAFE-CHAINN Ltd · Co. No. 12038453 · samantha@safe-chain.org · safe-chain.org · © 2026 Samantha Avril-Andreassen FRSA. All rights reserved.
The Directive™ · Policy Analysis · June 2026 · Not legal advice · Policy analysis and structural reform proposals in the public interest.