JusFund AI Ethics Policy

The platform uses AI at five points, in the same order for every case.

Reading. Every page of the uploaded record is extracted and classified from its content. Scanned pages are read by a document-reading service. The record is treated as untrusted input: instructions found inside a document are never treated as instructions to the platform.

Research. The parties, claims, stage and governing law are identified, and trusted public sources for the specific case are gathered and stored with the assessment. Only sources in that stored set can be cited.

Scoring. The record is first argued from the respondent's side. Eight assessment areas are then reasoned in writing, with the evidence for and against each. Versioned code, not the model, performs every calculation: the weighting, the adjustments, the caps and the final classification into a band with a percentage range and a confidence level.

Verification. Citations and quotations are checked against the record and the stored sources, the arithmetic is recomputed, and a second AI model from a different provider reviews the reasoning for contradiction, overstatement and unsupported conclusions. Its verdict stays attached to the report. A failed check is reported, not hidden.

Anonymisation. When a fund seeker lists a case, the platform drafts an anonymised summary and a second model checks it for anything that could identify the parties. The fund seeker reviews it before anything is shown.

The AI does not decide anything. It does not decide whether a case is listed, whether a funder is admitted, whether access is granted, or whether a case is funded. Those decisions are taken by people: the fund seeker, LexQuity staff and the funder.

The AI does not give legal, financial or investment advice, does not predict what a tribunal will decide, and does not produce a valuation. An assessment describes the documentary strength of a record on a published framework at a point in time. A percentage is never shown without its range and confidence level, and the framework is not statistically calibrated against concluded outcomes. Where the record does not support a responsible assessment, the platform declines to score it and says what is missing instead.

The AI does not learn from your case. No document, assessment or conversation on the platform is used to train any model, ours or a provider's. A case is never used as reference material for another case.

The AI does not profile people. It does not infer protected characteristics, does not score individuals, and is not used to make automated decisions with legal or similarly significant effects.

Every assessment goes to the fund seeker first. The reasoning behind every figure, the sources relied on, the limitations and the information request are part of every report, so a reader can check the work rather than take it on trust. You can question any output, add material and obtain a new version, and ask LexQuity for a human review of an assessment by writing to info@jusfund.tech. Every version is kept, every assessment records the framework version used, and the verification verdict stays visible.

The platform sends the text of a record, section by section, to a small number of AI model providers, and sends the finished assessment to a second review model from a different provider. It sends queries derived from the facts of a case, never the documents, to search services. Before any provider is used, LexQuity checks its security, its retention practices, its training practices and its sub-processors, and contracts on terms that prohibit training on our users' content and limit retention to what the service needs.

We do not publish the names of the model providers, because they change as the platform is tuned, and naming one would date this policy. We do publish where they are. The record is stored in the European Union. Model providers used for reading and scoring are established in the United States and the European Union. The independent review provider's international service operates from outside the United Kingdom and the European Union. Each transfer of personal data outside the United Kingdom or the European Economic Area is made under the safeguards the Privacy Notice describes, with a transfer risk assessment on file, and a provider that cannot meet those terms is not used, whatever its technical merit. The current list of providers and locations is available on request.

We test the framework for internal consistency and for disparities that would be unjustified: whether outputs vary systematically with the seat, the institution, the language or style of the documents, or the type of party. Apparently neutral features can stand in for protected characteristics, and where a feature does that it is removed or controlled. We monitor for drift when a model, a prompt or a provider changes, and a material change triggers renewed validation before it reaches a user. We publish what we measure and what we do not yet know, including that calibration against a representative set of concluded cases has not yet been done.

Uploaded documents are encrypted in transit and at rest and held in private storage. Access to a document, a report or a listing is role-based and recorded. AI processing of a case is logged: which documents, when, which stage, which provider category, and what was produced. Prompt injection through uploaded documents is treated as an attack and designed against. Any incident in which AI processing compromises confidentiality, accuracy or fairness is handled under our incident procedure, and where personal data is involved, under the breach rules of the Privacy Notice.

Every AI-generated output on the platform is labelled as such. Every report shows the framework version, the verification verdict and the limitations that apply. Nothing generated by AI is presented as written by a person, and no output is presented as more precise than the analysis supports.

European Union. LexQuity has assessed its use of AI against the EU Artificial Intelligence Act. The platform does not carry out a prohibited practice. It is not used for a purpose listed as high-risk: it does not assess natural persons for credit, employment, education, essential services, law enforcement, migration or the administration of justice, and arbitrators never receive a band. The transparency duties that apply to AI-generated content are met by labelling and by this policy. The assessment is kept under review as the Act's obligations phase in and as the platform changes.

United Kingdom. The platform is operated in line with the principles of safety, transparency, fairness, accountability and contestability that UK regulators apply, and with the UK GDPR's rules on automated decision-making, which the platform does not engage.

United States. The platform is not used for any consequential decision about an individual, and its outputs are labelled and explained. Where a state law on automated decision tools applies to a use of the platform, that use is assessed before it is offered.

Singapore. The platform follows the Model AI Governance Framework's principles of explainability, transparency, fairness and human centricity, and the PDPA's rules on the personal data the AI processes.

United Arab Emirates. The platform follows the UAE's AI ethics principles of fairness, accountability, transparency, explainability and privacy, and the Personal Data Protection Law on the data involved.

Everywhere. Where a law where you are imposes a stricter requirement, that requirement prevails.

This policy is reviewed at least once a year, and sooner after a material change to the platform, to a provider, or to the law. Questions and challenges go to info@jusfund.tech.

In short: the AI reads, researches, reasons and checks. Code does the arithmetic. People decide. Nothing you upload teaches a model, and nothing about your case reaches a funder unless you say so.