Processing delays for disability requests

In 2024, 1.8 million people submitted requests to a departmental house for people with disabilities (MDPH), and the average processing time reached 4.8 months, according to the Key Figures on Autonomy Assistance 2026 from the National Solidarity Fund for Autonomy (CNSA).

AI experiment in MDPHs

The CNSA conducted an experiment in 11 MDPHs using AI to handle initial stages of processing requests, such as splitting scanned files, recognizing documents, and checking if the requests were complete. The experiment also included a legal chatbot, with results expected by mid-October 2026.

Results not yet published

As of October 20, the CNSA had not published the results of the experiment or the evaluation method. The fund is in the final phase of its AI Roadmap 2025-2026, with 29 of the 36 planned actions already deployed.

Test framework and limitations

The test framework, a public document from the CNSA, limited the AI to paper files and required deceased persons' files as test data. It excluded any writing in the MDPH information system, focusing on the AI's ability to sort real pieces of documents.

Guarantees and selection process

The guarantees for using AI on ongoing requests, such as acceptable error rates and correction of wrongly rejected pieces, remain to be set. Up to ten MDPHs, 250,000 euros, and a jury to evaluate tools were part of the selection process.

Legal and technical requirements

The framework required tools to be hosted in France or Europe and not subject to the Cloud Act or the Patriot Act. It also specified that a legal chatbot, based on generative AI, must cite sources precisely and not replace human expertise.

Document management and checks

The second use case involved a tool that must separate scanned paper files into documents, recognize each piece, name it, and classify it. It checks if the request is complete and extracts key information like the applicant's identity and submission date.

Ethical and legal considerations

The test framework established a clear ethical and data security framework with mandatory human supervision. However, it did not specify success thresholds or data protection impact analysis, which the GDPR requires for large-scale health data processing.

Generalization and future steps

The CNSA announced that if the results are conclusive, the AI tools may be generalized across the entire territory. However, no schedule or criteria for generalization were provided, and the test book and indicators were not published.

Comparison with other regions

In Bavaria, the Landratsamt of Kelheim introduced a pilot project using AI to verify documents during the online application for residence permits. Unlike the French test, this AI alerts applicants during the process rather than assisting agents with already received files.