French local authorities are no longer just digitizing forms. Since 2026, a structured co-financing mechanism allows municipalities, inter-municipalities, and departments to deploy sovereign artificial intelligence solutions on a large scale. This paradigm shift, from isolated pilots to continuous funding, reshapes how local public services integrate technological innovations for the benefit of citizens.
AI Territories Desk: the funding criteria that guide projects
Since May 21, 2026, the Banque des Territoires has opened a permanent desk called AI Territories. This system covers 30 to 50% of the cost of sovereign AI solutions deployed by territorial structures. No application deadline has been set: the desk remains accessible for three years.
What distinguishes this mechanism from previous calls for projects is the criteria grid that conditions the level of coverage. A co-financing bonus is granted to projects shared among several local authorities, to territories classified as fragile, and to solutions qualified as frugal AI. The overall ceiling reaches 50% coverage for projects that meet these criteria.
In practice, this bonus logic encourages local authorities to pool their needs rather than launch siloed experiments. A rural inter-municipality that shares a road degradation detection tool with three neighboring municipalities accesses a higher funding rate than a metropolis acting alone. Several resources document the modalities of this system, notably the website cc-paysdebecherel.fr which lists territorial technological initiatives.
| Bonus Criterion | Effect on Co-Financing | Target Type of Local Authority |
|---|---|---|
| Pooling among local authorities | Increase in rate towards the 50% ceiling | Inter-municipalities, mixed unions |
| Fragile territory | Dedicated bonus, priority access | Rural municipalities, mountain areas |
| Frugal AI (low footprint) | Additional bonus | All structures |
| Large user base | Increase in rate | Departments, large urban areas |
Training Public Agents in Artificial Intelligence: the Goal of 100,000
Technological deployment only produces results if the agents using the tools master them. The national program aims to train 100,000 territorial public agents in AI by 2030. This figure represents a scaling up compared to the occasional awareness sessions organized so far.
However, small local authorities struggle to free up agent time for training. Town hall secretaries, often the only ones in position, do not have the same leeway as a digital service in a metropolis. The training plan will need to take this asymmetry into account to achieve its goal.
AI Assistants in Town Halls: Concrete Data on Citizen Use
The chatbots deployed in municipalities provide measurable indicators. In Plaisir, in the Yvelines, the conversational agent Optimus reduced the lost message rate from 65% to 8%. In Issy-les-Moulineaux, IssyGPT handled nearly 15,000 requests in less than a year.
These figures deserve careful reading. A lost message does not mean a citizen request was ignored: it can be duplicates, spam, or out-of-scope requests. The decrease remains significant and reflects better capture of legitimate solicitations.
- AI assistants prioritize recurring questions (hours, civil status procedures, waste collection), freeing up agent time for complex situations.
- Generative AI allows for personalized responses to the local context, unlike static FAQs that poorly cover the diversity of municipal situations.
- Continuous operation (24/7) meets observed usage: a notable share of requests comes in outside of counter opening hours.
Sovereign AI and Territorial Data Management: A Constraint Turned Lever
The AI Territories desk requires that the funded solutions rely on hosting and technologies controlled within the territory. This sovereignty requirement, initially perceived as a technical constraint, directs local authorities towards French or European providers capable of guaranteeing data localization.
For citizens, the question arises in terms of trust. The data processed by a municipal AI assistant (identity, address, family situation) falls within the most sensitive area of public service. Choosing sovereign hosting does not eliminate risks, but it places the responsibility for processing within a readable legal framework.
The challenge for local authorities is to move from awareness to operational action, with tools integrated into existing information systems rather than added as an overlay.
Structured funding, large-scale training, and the requirement for sovereignty outline a model where technological innovation in territories no longer depends on isolated initiatives. The data that summarizes this trajectory remains the maximum co-financing rate of 50%: high enough to trigger projects, conditioned on criteria that prevent dispersion.
