French Justice: An AI Assistant for Sovereignty

Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisis ton rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
French Justice: An AI Assistant for Sovereignty
The digitization of Justice is underway, according to Minister of Justice Gérald Darmanin. In an interview with Franceinfo regarding the shortcomings in the Lyhanna case, the minister unveiled his plan to digitalize the courts, with a target of achieving “zero paper” within six months.
This was an opportunity for the minister to remind everyone about the launch of a generative AI assistant, named “Mon Assistant Justice,” which was presented last May. Developed by the Interministerial Digital Directorate (Dinum), this tool is intended to enable thousands of agents to perform tasks such as writing, synthesis, translation, and ideation in an environment described as secure and sovereign.
Justice Facing the Limits of Existing Offers
According to Gérald Darmanin, the offerings currently available on the market do not meet the specific constraints of the Ministry of Justice, thus justifying the development of yet another tool. He mentioned the particularly sensitive nature of the data being processed, which relates to, for example, investigative secrecy, enquiry secrecy, or ongoing cases. This is why it is impossible “to put this data in clouds that are outside of France,” he argues.
It is worth noting that data can be hosted on French territory while still being accessible to foreign authorities when a provider is subject to extraterritorial legislation, such as the American CLOUD Act. This analysis reveals a common misconception in public debate: the risk does not solely depend on the location of the data but also, and especially, on the nationality of the cloud provider that processes it.
Sovereignty Does Not Replace Cybersecurity
The ministry thus claims to have chosen a tool hosted in France to avoid “uncontrolled uses of public tools” that could expose this data to “real risks of foreign interference or leaks,” as stated in official communications.
This last argument deserves nuance. While sovereignty can help mitigate certain risks associated with extraterritorial legislations, it does not in itself guarantee cybersecurity. Indeed, data protection primarily depends on the technical and organizational measures implemented by the administration and its service providers. Recent events serve as a good example of this.
For the Minister of Justice, market solutions do not fully meet the demands of the Justice system. “We do not have infrastructure today among the major legal digital publishers,” he explains, adding: “Imagine if we conducted searches at Airbus (...), took data and put it with the Americans [data hosted by an American cloud provider, ed. note], that would greatly interest Boeing.”
When the State Becomes a Software Publisher
As is the case with the “Suite Numérique,” a set of collaborative office tools, the State is gradually finding itself in a position that goes beyond that of a mere technology user. However, this strategy is not without its trade-offs. By gradually becoming the project owner, and sometimes even the project manager of its own digital tools, the State exposes itself to a recurring challenge: maintaining solutions that can keep pace with the innovation rate of a particularly competitive AI market.
Rather than replacing the market, the State could seek to express its requirements more clearly regarding sovereignty, security, and data governance to encourage providers to adapt their offerings. This strategy would allow it to benefit from the innovation pace of the private sector while retaining the level of control it deems necessary.
The Challenge of a Tool That Meets Expectations
It also requires the administration to demonstrate its ability to sustain these projects over time. The development of an artificial intelligence assistant is not limited to launching a first version. It entails continuously funding the evolution of features, improving performance, integrating new models, correcting identified biases, and adapting to user needs.
However, if the tool fails to compete with market offerings, agents may be tempted to turn to external solutions, thereby recreating the uncontrolled uses that the ministry is specifically trying to avoid.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.