Vision Pierre Genevier Version PDF

SUMMARY NOTE — PROPOSAL No. 3

Reform of legal aid — specialized corps of judges and lawyers, global management system, and AI assistance system

Document prepared with Claude (Anthropic) — version revised on 17 September 2026, incorporating comments submitted by Pierre Genevier on 17 September 2026 regarding the version of 11 August 2026.

Living working document — the figures and assumptions presented here constitute an initial estimate, explicitly intended to be refined in a forthcoming session.

1. Purpose and positioning within the platform

This note addresses Proposal No. 3 of Pierre Genevier's platform: the reform of the legal aid system, based on the creation of specialized corps of civil-servant judges and lawyers, the development of a global system for managing and classifying cases, and an artificial intelligence system assisting both corps as well as legal-aid applicants.

In accordance with the common template validated on 15 July 2026, this note focuses on what is specific to Proposal No. 3 and refers to Summary Note — Proposal No. 1 for everything relating to the IO's common institutional architecture, its infrastructure scenarios (Scenarios A, B, C) and its overall financing as an institution. It draws in particular on the legal-aid AI system development timeline already established on 14 July 2026, which it supplements here with a cost estimate that had until now remained unquantified.

Method note: this version carries forward the content of the 11 August 2026 version and incorporates the elements submitted by Pierre Genevier on 17 September 2026, namely: a new feature allowing occasional recourse to guest independent lawyers, as backup support for civil-servant lawyers (sections 2.3, 3.1 and 6), with the corresponding effects on the costing of Block A (sections 3.1, 4.1 and 4.4); alignment of the heading of module 2 in the table in section 3.1 with that of module 3, to reflect that the Legal-Aid Judges Application also records time and costs (with no impact on the person-month estimate); precise verification and sourcing of the French statistics, based on the parliamentary reports Joissains-Mézard (Senate, 2014) and Gosselin-Moutchou (National Assembly, 2019), correcting the rate from 24% to 22% (sections 3.3, 6 and Annex B); and the addition of the example of Japan (Japan Legal Support Center / Houterasu), retained both as a precedent strengthening the acceptability of turning lawyers into civil servants and as a candidate, alongside France, for the list of pilot States (sections 3.3 and 6). The costing of expected savings (section 4.3) remains deferred to a forthcoming session, pending further input. It remains a living working document, to be enriched in future sessions.

2. Description of the proposal

2.1 Three layers to distinguish

The proposed reform combines three layers of a different nature, which should be distinguished for the remainder of the note:

• an institutional reform, specific to each participating State: creation of a corps of civil-servant judges specialized in examining legal-aid applications, and a corps of civil-servant lawyers specialized in legal-aid assignments, both placed under the hierarchical responsibility of the State and of the UN Office of the High Commissioner for Human Rights (OHCHR);

• a management and data layer: global and common to participating States — international classification and coding of case types, databases of adjudicated cases and legal-aid parties, recording of time and costs;

• an artificial intelligence layer: building on the previous two — decision support for legal-aid judges, drafting and defense assistance for legal-aid lawyers.

The three global applications described in section 2.3 (judges application, lawyers application, applicants application) are the shared technical backbone for the second and third layers; they carry both the management/data mechanics and the AI that assists them. The feasibility of these two layers is therefore addressed separately in section 3 (Block A — management system; Block B — AI system), while the institutional reform itself (first layer) is a matter for each State's own political and legislative process, outside the technical scope of this note.

2.2 Institutional reform: a single national legal-aid office (BAJ)

The mechanism rests on a single national legal-aid office (BAJ) per participating State, bringing together judges specialized in examining legal-aid applications and in mediation. These judges would investigate applications (gathering documents and information, hearing the parties), attempt to settle cases amicably through mediation where possible, and follow cases throughout the proceedings — the same judge being able to rule on the legal-aid application at first instance, then for any appeal, any cassation appeal, and any application to an international or regional human-rights court.

The expected benefits of this organization include: a reduction in the workload of courts through greater recourse to mediation (22% of cases judged each year in France involve at least one party benefiting from legal aid); a unified working methodology, supported by advanced technologies; a simplification of the examination of applications at higher levels of jurisdiction, since the judges will already have investigated the file at lower levels; a reduction in the operating cost of BAJs through the pooling of management and IT expenditure among participating States; and an improvement in the quality of legal-aid decisions, reducing both the risk of lost fees for the State and the risk of a poor person having to defend themselves alone.

A cross-cutting benefit, particularly relevant to this note, concerns the improvement of the information system on legal aid and on justice in general: the reform would, for the first time, make it possible to establish the real total cost of the legal-aid system (unknown today, including in France) and its breakdown by cost category (management, fees, travel), as well as better coordination with the existing information systems of Ministries of Justice.

2.3 Corps of civil-servant lawyers specialized in legal-aid assignments

The creation of a corps of civil-servant lawyers specialized in legal aid would aim to guarantee respect for the constitutional rights of poor persons — not systematically ensured today — while minimizing total legal-aid and management spending, and maximizing the use of public-spending-reduction mechanisms.

This respect for constitutional rights would rest notably on: a unified working methodology and a quality-control system, supported by the IT system described in section 2.4; supervision of lawyers with less than five years' experience by an experienced lawyer; assignment of the most difficult cases to the most experienced lawyers, with several levels of remuneration according to skill — impossible under the current French system; a precise time-per-case-type scale, taking into account the lawyer's skill and the difficulty of the file; a reduction in conflicts of interest linked to the use of independent lawyers; and the elimination of upfront costs borne by lawyers, who would now be paid regularly.

Spending optimization would rest notably on: the ability to precisely establish the total cost of legal aid, item by item — an assessment the Cour des comptes already recommends carrying out; strengthened pooling of management and IT expenditure, notably via the applications described in section 2.4, AI, and videoconferencing to communicate with courts and prisons; revenue generation on certain cases (a percentage of damages awarded, along the lines of certain US mechanisms) and greater use of legal-aid reimbursement mechanisms from the losing party; and a simplification of lawyer payment — in France, moving from around 1 million legal-aid assignments paid to more than 25,000 lawyers, to around 12 annual payments to around 8,000 lawyers.

Addition of 17 September 2026 — Occasional recourse to independent lawyers (backup mechanism): in order to absorb workload peaks — particularly during the ramp-up phase of the system, when experience does not yet allow the required number of civil-servant lawyers to be precisely calibrated (cf. section 3.3) — the mechanism provides for the possibility of engaging an independent lawyer on a specific legal-aid case. That lawyer is granted guest access, limited to the case concerned, to the Legal-Aid Lawyers Application (section 2.4), so that case documents are kept in the common system and the working methodology (time scale, quality control) continues to be respected even outside the civil-servant corps. This mechanism is a flexibility valve rather than a permanent alternative pathway: it addresses a transitional sizing need, not a challenge to the principle of a corps of civil-servant lawyers.

2.4 The three global applications and the data layer

The technical system rests on three global internet applications, intended to be used by all participating States:

• “Legal-Aid Judges” Application: management and examination of legal-aid applications, mediation support, case tracking, recording of time spent and costs associated with examining each application.

• “Legal-Aid Lawyers” Application: management of defense files for poor persons, recording of time spent and costs associated with each file.

• “Applicants” Application: assistance for poor persons in formulating their legal-aid application.

These three applications would also make it possible to build: an international classification and coding of all types of cases brought before the courts, in each country and each year; a database of adjudicated cases and a database of legal-aid parties, including offenders and criminals — useful for monitoring the work of legal-aid judges and lawyers as well as for combating organized and transnational crime and terrorism; and management data useful for improving justice and police systems. This cross-cutting data action falls within the UN Data Strategy (reference PJ No. 55 of the candidacy file).

Point of caution to note at this stage and revisited in section 6: the database of legal-aid parties, in particular its component relating to offenders and criminals, touches on sensitive criminal data and directly connects with the question of personal-data governance already identified in Summary Note — Proposal No. 1 (section 5.4 of the consolidated note of 14 July, on the phasing between legal entities and natural persons).

3. Technical feasibility

Technical feasibility is addressed in two distinct blocks, in accordance with section 2.1: Block A covers the management system (the three applications and the data layer, excluding AI); Block B covers the AI assistance system, which builds on Block A once the latter is operational.

3.1 Block A — Management system (judges, lawyers, applicants)

By analogy with the method already applied to Proposal No. 2 (domain-name pricing engine), the management system is broken down into functional modules, with an indicative estimate of complexity and development effort — first-pass assumptions, to be refined once the technical architecture is validated and the development staffing is finalized.

Module 7 (national interconnection) differs from the other seven in its recurring nature: each new State joining the system entails its own integration effort, whereas modules 1 to 6 and 8 constitute a global core to be developed only once. The total below therefore covers only this global core, excluding the per-State integration cost.

Indicative total for the global core (modules 1 to 6 and 8): on the order of 66 to 107 person-months (including the guest-access profile for independent lawyers, module 3). With a team of 8 to 14 people (developers, data engineer, comparative legal taxonomy expert, security expert, project manager), this corresponds to a parallel development schedule of 12 to 18 months for a first complete version of the core, excluding national integration (module 7) and excluding large-scale testing.

Recommended phasing, on the same principle as Proposal No. 2: an MVP limited to a small number of pilot States (modules 1, 2 and 3 over a reduced scope), an extension phase (module 4 and generalization of the classification), then a consolidation phase (modules 5, 6 and 8, including the legal-aid parties database and its enhanced security requirements).

3.2 Block B — AI assistance system

The AI system's development timeline, differentiated according to the three infrastructure scenarios defined in Summary Note — Proposal No. 1, was already established on 14 July 2026. It is recalled below and supplemented with an initial cost estimate, previously absent.

The staffing and development-effort figures given here are first-pass orders of magnitude, distinct from the infrastructure costs (CAPEX/OPEX) already costed in Summary Note — Proposal No. 1: they correspond to the cost of the team developing the AI system itself (corpus, fine-tuning, specialization, testing, deployment), not to the cost of the data centers on which that system runs.

Moving from Scenario A to Scenario B does not mean starting from scratch for legal aid (cf. the 14 July note, section 4.3): it involves either continuing to fine-tune an open-weights model on the IO's own infrastructure, or engaging a moderate-size general-purpose model, with the jurisdiction-specialization pipeline (corpus, testing, deployment) remaining largely reusable. Scenario B's development effort therefore includes a share of work carried over from Scenario A, not deducted in the table above as a matter of methodological caution.

3.3 Volumetry

The data submitted for France, now verified and sourced from two parliamentary reports (cf. Annex B), provide an initial benchmark: 1,132,581 legal-aid applications filed in 2017, of which 985,110 were granted and 79,625 rejected; around 22% of cases judged each year involve at least one party under legal aid (915,563 legal-aid grants against 4,128,472 cases concluded in 2013); around 4.1 million cases judged in total each year.

Unlike Proposal No. 2, whose global volumetry can be approximated from already-available global data (number of businesses, number of domain names), the volumetry of Proposal No. 3 depends directly on the number and profile of States participating in the system — the institutional reform described in section 2.2 not being intended to be imposed but adopted voluntarily by each State. A global extrapolation by a simple demographic rule of three (world population / French population) would therefore have only very limited illustrative value, as rates of recourse to justice and legal aid vary considerably according to legal systems and income levels.

It is therefore recommended, at this stage, to use the French figures as the reference unit per State, and to defer the global costing until an indicative list of pilot States is set — a logic already adopted for infrastructure Scenario A (“initial pilot with a limited scope”, Summary Note — Proposal No. 1, section 3.1).

Addition of 17 September 2026: on this basis, France and Japan constitute two natural candidates for an initial scope of pilot States. France has the most complete and best-sourced data at this stage (cf. Annex B), making it a calibration benchmark for volumetry and costing. Japan, for its part, already has a corps of salaried lawyers dedicated to legal aid within its public legal-aid body (Japan Legal Support Center / Houterasu), making it an operational precedent directly relevant to the institutional reform described in section 2.2 and to the acceptability of turning lawyers into civil servants, addressed in section 6.

3.4 Infrastructure and security needs

Block A, like Proposal No. 2, is a matter of a transactional database rather than a frontier AI computing need: the volumes involved (a few million cases per year and per State, with history) remain of a manageable order of magnitude without massive dedicated infrastructure.

Clarification added on 11 August 2026: this assessment is based on the French example and constitutes an average, which masks potentially significant heterogeneity among States. Highly populated States with an active judicial system — India being the clearest example, with a volume of cases judged each year far greater than France's — could on their own represent a significantly higher load than the average used here, while a large number of smaller or less litigious States will remain well below a million cases per year. The assumption of an order of magnitude “manageable without massive dedicated infrastructure” remains reasonable on average and for a shared global core, but will need to be revised upward for the sizing specific to each pilot State once the list of participating States is set (cf. section 3.3) — particularly if very high-volume States join the system as early as the pilot phase rather than at a later stage.

The main point of caution concerns security rather than volume: the legal-aid parties database (module 6), including data on offenders and criminals, calls for a level of security and access control comparable to that of an international criminal record, with data-sovereignty governance still to be decided (centralized hosting vs. regional distribution), on the same model as the question already raised for Proposal No. 2 (section 5.2.4 of the 14 July note).

4. Financial feasibility

4.1 Cost of Block A (management system)

On the same principle adopted for Proposal No. 2 (section 4.2 of the 11 August 2026 note), Block A is a large-scale transactional application carried by an international organization. However, the enhanced security level required by module 6 (criminal data, comparable to an international criminal record) justifies using a range slightly higher than that of Proposal No. 2: an average fully loaded cost on the order of $20,000 to $30,000 per person-month, all-inclusive (loaded salary, social contributions, project management, overheads), excluding infrastructure.

Applied to the estimates in section 3.1, this assumption gives:

• Global core (modules 1 to 6 and 8, 66 to 107 person-months): estimated development cost of between $1.32M and $3.21M (versus $1.26M to $3.09M before the addition of the guest-access profile for independent lawyers).

• Integration per pilot State (module 7, 12 to 20 person-months per State, in parallel): estimated cost of between $0.24M and $0.60M per State, to be multiplied by the number of States joining the system — not yet totalable as long as the list of pilot States is not set (cf. section 3.3).

4.2 Cost of Block B (AI system)

Developing a specialized AI system draws on profiles (machine-learning engineers, model-specialization specialists, domain legal experts) whose average cost on the global market is structurally higher than that of a classic transactional-application development team. A range of $28,000 to $40,000 per person-month is therefore used as a working assumption for Block B — to be distinguished from infrastructure cost (CAPEX/OPEX), already costed overall in Summary Note — Proposal No. 1, and from model-training costs proper, not included in this range.

Applied to the estimates in section 3.2, this assumption gives:

• Scenario A (60 to 110 person-months): estimated development cost of between $1.68M and $4.4M.

• Scenario B (300 to 550 person-months, excluding model-training costs): estimated development cost of between $8.4M and $22M.

These amounts remain, at this stage, first-pass orders of magnitude, to be refined once the technical architecture and staffing are finalized, and to be weighed against the expected savings from the institutional reform (section 4.3) once those are documented. The proportional share of AI infrastructure dedicated to legal aid within the overall CAPEX/OPEX of Scenarios A and B (recalled in section 3.2) remains, as noted previously, to be refined separately — a simple rule of proportion to the number of data centers gives an order of magnitude but probably not a precise estimate.

4.3 Expected savings from the institutional reform

The institutional reform described in sections 2.2 and 2.3 aims, beyond the development cost of Blocks A and B, at a reduction in public legal-aid spending in each participating State, through several channels: pooling of management and IT expenditure among States; reduced recourse to courts through mediation; revenue generation on certain cases; and simplification of lawyer payment (moving from per-assignment payment to a salary system). These savings are not costed in this version of the note, owing to a lack of sufficiently reliable data on current costs — the note itself points out that the total current cost of the legal-aid system is not known with precision today, including in France. Costing these savings therefore presupposes, as a preliminary step, establishing the system's current cost through the mechanism itself (section 2.2), which creates a methodological dependency to be documented in section 6. There is no new development on this point at this stage; it will be revisited once the other parts of the note are more advanced, in line with the exchange of 11 August 2026.

4.4 Summary

5. Implementation timeline

• Phase 0 — Institutional reform: a political and legislative process specific to each pilot State, a prerequisite for the operational launch of the single national BAJ; outside the technical scope but conditions the overall timeline.

• Phase 1 — Block A, MVP (modules 1 to 3 over a pilot scope): in parallel with infrastructure Scenario A defined in Summary Note — Proposal No. 1.

• Phase 2 — Block B, Scenario A (9 to 15 months): AI system on external inference/fine-tuning, building on the Block A core once the MVP is available.

• Phase 3 — Block A, extension and consolidation (modules 4 to 8, then progressive integration of States joining the system): in parallel with the move to infrastructure Scenario B.

• Phase 4 — Block B, Scenario B (24 to 36 months): migration to infrastructure owned by the IO, multi-jurisdictional and multilingual extension.

• Phase 5 — Block B, Scenario C (conditional): migration of the existing pipeline to the IO's own frontier model, according to the Scenario C timeline arbitrated at the level of the IO as a whole (Summary Note — Proposal No. 1, section 5).

This timeline remains, like that of Summary Note — Proposal No. 1, to be specified year by year in a forthcoming session, in particular to determine the number and identity of the Phase 1 pilot States — a parameter on which volumetry (section 3.3) and the national integration cost (section 3.1, module 7) directly depend.

6. Risks and open points

• French figures: verified and sourced on 17 September 2026 from two parliamentary reports (1,132,581 legal-aid applications filed in 2017, of which 985,110 granted and 79,625 rejected; 22% of cases judged each year involving at least one party under legal aid; around 4.1 million cases judged per year) — see Annex B for details and complete references. Point remaining open: these figures describe the French situation and do not prejudge the volumetry of other potential pilot States (cf. section 3.3).

• Global volumetry: depends on the number and profile of the pilot States, not yet fixed — a point to be settled before any global costing (section 3.3). Point of caution added on 11 August 2026: the assumption of a volume “manageable without massive dedicated infrastructure” (section 3.4) is an average built on the French example and will need to be revised upward for very high-volume States (India in particular), where applicable as early as the pilot phase.

• Average cost-per-person-month assumptions ($20,000–30,000 for Block A, $28,000–40,000 for Block B, sections 4.1 and 4.2): indicative orders of magnitude, not yet validated by Pierre Genevier.

Political acceptability of turning specialized legal-aid lawyers into civil servants: the reform affects the status of an organized liberal profession (bar associations) and could give rise to resistance that should be documented and anticipated in the candidacy strategy, including in France. Nuance added on 11 August 2026: this risk must be weighed against the argument that, if the public AI assistance system (Block B) proves effective, the expected gains for justice, society and poor persons would be considerable — an argument to be developed, where relevant, in the communication and negotiation strategy with the professions concerned, without this exempting the note from documenting the resistance itself. Additional nuance added on 17 September 2026: the mechanism for occasional recourse to guest independent lawyers (section 2.3) partly addresses this risk by offering, from the outset, a pathway for collaboration for non-civil-servant lawyers, which limits the all-or-nothing character of the reform in the eyes of bar associations; moreover, the example of Japan, where the Japan Legal Support Center (Houterasu) already employs salaried lawyers dedicated to legal aid, constitutes a real precedent showing that a corps of civil-servant lawyers specialized in legal aid can work, making it both an acceptability argument and a relevant candidate for the list of pilot States (cf. section 3.3).

• Articulation of the dual State/OHCHR oversight of the legal-aid judges and lawyers corps with the IO's institutional architecture (legal status and internal governance still to be developed — Summary Note — Proposal No. 1, section 6): to be clarified.

• Sensitivity of the legal-aid parties database (offenders, criminals): directly connects with the still-open question of phasing between legal entities and natural persons, and the federated model recommended for individual data (Summary Note — Proposal No. 1, section 5.4 of the 14 July note).

• Methodological dependency of the costing of expected savings (section 4.3) on first establishing the real current cost of the legal-aid system, which the reform itself aims to produce — to be documented as an accepted limitation rather than minimized, on the model already adopted for the self-declaration of Internet-specific data in Proposal No. 2. No new development at this stage (exchange of 11 August 2026); to be revisited once the other parts of the note are further advanced.

• Recurring national integration cost (module 7, Block A): not included in the global-core total; to be costed once the list of pilot States is set.

• Proportional share of AI infrastructure dedicated to legal aid within the overall CAPEX/OPEX of Scenarios A and B: estimate by simple proportion to the number of data centers judged approximate, to be refined.

• Cost of Block B under Scenario C: not costed at this stage, pending a finalized timeline for this scenario (already identified as an open point in Summary Note — Proposal No. 1).

7. Annexes

7.1 Annex A — Detail of Block A modules (management system)

Reproduced from section 3.1 for quick reference.

7.2 Annex B — French reference data (verified and sourced on 17 September 2026)

• Legal-aid applications (2017 data): 1,132,581 applications filed, of which 985,110 granted and 79,625 rejected (source: Gosselin-Moutchou report, National Assembly, 23 July 2019, p. 13, which also details lapsed applications and other grounds for closure).

• Cases judged involving a party under legal aid (2013 data): 22% of cases judged each year, i.e. 915,563 legal-aid grants against 4,128,472 cases concluded (source: Joissains-Mézard report, Senate, 2014, p. 17).

• Total cases judged (2013 data): 4,128,472 per year (same source, p. 17).

• Legal-aid spending per capita in Europe: comparative data available in the 2019 report, p. 20 — not yet used in this note, to be examined for an international comparison in a forthcoming version.

• Order of magnitude of the legal-aid lawyers corps after reform (France): around 8,000 civil-servant lawyers, compared with more than 25,000 lawyers currently assigned cases.

• Sources: Joissains-Mézard report (Senate, 2014), http://www.pierregenevier.eu/npdf2/rapport-AJ-joissains-7-2014.pdf; Gosselin-Moutchou report (National Assembly, 23 July 2019), http://www.pierregenevier.eu/npdf2/rap-AJ-Moutchou-23-7-19.pdf.

These figures, initially submitted on 16 July 2026 as estimates, were verified and sourced by Pierre Genevier on 17 September 2026 by direct reference to the two parliamentary reports cited above.