AI support to dense data court applications

This article considers how available AI programs can support the preparation of judgments in document heavy cases, with implications for judges and those supporting them in organising documents and preparing submissions.  We look at how programs and The Judiciary’s 14 April 2025 AI Guidance (“April 2025 Guidance”) could have applied to the well-known case of Avanti.

The Avanti restructuring

Avanti was a satellite communications business.  It had connected-party debts approaching US $1 billion and non-connected unsecured debts of over £30 million. The secured creditors claimed fixed charges over the company’s core assets and the total asset value fell significantly short of their debt. The business was restructured through a pre-pack administration. In Avanti Communications Ltd [2023] EWHC 940 (Ch), Johnson J considered whether the security created over certain key assets including satellites, ground-station equipment, International Telecommunication Union (“ITU”) filings and Ofcom licences – constituted fixed or floating charges. The characterisation was critical to the distribution of sale proceeds.

The problem of dense security documentation

The judgment shows the difficulties that can arise in multi-instrument cases such as finance structures, particularly where different definitions and covenants apply in documents created at different times for different situations. The court heard extensive submissions from leading counsel on both sides, working through hundreds of pages of finance agreements, security documents and carve-outs. The judge referred to the material as a “thicket of contractual provisions”.

Issues arising included the considerable number of documents, the contractual restrictions on Avanti’s right to manage its assets, and the apparent absence of the key definition “Debt Documents” from the most recent debenture.  The judge had to assume it mirrored the “Note Documents” in an earlier debenture. The judge expressly acknowledged the limitations of his summary of the documents, noting that it was not “a detailed and comprehensive summary of all the relevant provisions” and recording his indebtedness to counsel for guiding him through the material.

This candid recognition of the challenges involved highlights the cognitive demand on valuable judicial capacity.  We suggest potential uses for judge-in-the-loop AI. This support does not replace judicial reasoning but can assist in mapping and organisation of documents and issues.  

Where could AI have helped?

The April 2025 Guidance confirms that judges may use AI to summarise, map and cross-reference within closed document sets, provided outputs remain verifiable and auditable.  AI cannot properly be used for legal advice or substitute judicial reasoning.  In Avanti, AI could assist by:

1.  Mapping defined terms across instruments

AI could have flagged this missing definition in the 2027 debenture, highlighted its absence and proposed possible parallels from other instruments. That would have allowed the judge and counsel to test the drafting lineage directly, rather than proceeding on assumption.

2. Visualise each disposal path

The super-senior facility agreement permitted certain disposals, but subject to conditions and repayment obligations that, in the judge’s words, made transactions “commercially unattractive”.  AI could describe which disposals required lender consent, which demanded certificates or legal opinions, and which imposed economic deterrents, such as a 101% repayment waterfall. This would help draw distinctions between legal restriction and commercial friction.  A chart of these distinctions could assist in the key issue in the case – the characterisation of charges according to what legal rights were created over assets at the time the debentures were entered into.

3. Asset-class control matrix

Avanti’s assets were varied in type and use.  There were contractual restrictions on use and assets like orbital slots and spectrum licences were subject to additional external regulatory approvals.  AI could classify the assets from the date and summarise the controls in a simple matrix. That visualisation would clarify how the origin and effects of restrictions to assist the characterisation analysis.

4. Highlight absent stakeholder perspectives

The judge noted that unsecured creditors did not appear at the hearing and decided that this did not matter.   A simple AI prompt could supply a note of why they may not be there, what evidence they might refer to or what arguments they might have raised. This would be speculative but could in an appropriate case improve transparency by recording or directing enquiries into the counter-case.

Taken together, these modest interventions would not alter legal reasoning. They could provide judges and counsel clearer tools for navigating dense security packages, surfacing hidden ambiguities and ensuring the judgement records the full landscape before applying established case law.

Practical tools available to judges

• Microsoft Copilot (Judicial Edition) – now available on judicial devices via e-Judiciary, offering a secure, enterprise-level tool for summarisation and clause mapping within the judicial suite.

• Case management AI plugins – such as Lexis+ AI or tools embedded in case bundles, for clause extraction, summarisation and definition mapping.

• Diagramming & data visualisation AI – for transforming clause logic and security pathways into flowcharts and metrics – using secure platforms like Power BI Copilot or Visio Copilot.

Risks identified in the April 2025 Guidance

• Hallucination risk – AI might fabricate case law or definitions if not strictly confined to verified documents.

• Bias and data drift – particularly pertinent in interpreting non-standard or transnational covenants under English law.

• Confidentiality breaches – the April 2025 Guidance warns against using public AI chatbots for judicial or legal materials.

• Accountability and over-reliance – judicial office holders remain personally responsible for all AI-assisted material and must verify outputs; reasoning must be demonstrably theirs.

Training and implementation considerations

• Tooling costs – Judicial Copilot licences and secure integration; cost is estimated in thousands, not millions of pounds.

• Training – Short (half-day) courses or home e-learning packages covering:

  > What AI can/cannot do under guidance.

  > How to verify outputs and prevent hallucination.

  > Bias awareness and transparency in judgments.

• Self-training – Judges using non-confidential or publicly available sample cases to practice building definitional maps and flowcharts.

Why this matters for future cases

Avanti’s “thicket” of covenants and definitional gaps consumed valuable judicial time and cognitive effort. With proper use of AI tools:

• Latent ambiguities (e.g., missing definitions) would surface early.

• Control analysis could distinguish clauses creating relevant property rights from those promoting commercial deterrence.

• Transparency in dealing with potential stakeholder interests could improve.

• Judgment writing would be faster, while preserving judicial independence and fairness.

The April 2025 Guidance underscores that AI supports and does not replace judicial reasoning. Used securely and with judicial oversight, AI could help find a clearer passage through the thicket of data in cases.  Counsel may propose agreed protocols: confining AI to the case bundle, disclosing use of AI in visual aids or other submissions.  Litigators must note hallucination and confidentiality risks.  Early agreement on AI use will promote efficiency and transparency without compromising fairness.

Co-authors: Mark Parkhouse and Don Williams are post-grad students on Southampton University’s MA in Artificial Intelligence. Mark is a solicitor advocate and a partner at Aria Grace Law CIC. They were assisted in preparing this article by Sarah Davies, a trainee solicitor at Aria Grace Law CIC.