Mistral should not influence “the deployment choices” of its technology by the army, Arthur Mensch, boss of the French artificial intelligence start-up, told AFP on Thursday, at a time when ethical concerns on this subject are agitating the tech sector across the world.
“The choices of deployment and use are not our responsibility,” he declared during an interview on the sidelines of the first conference on artificial intelligence (AI) organized by his company in Paris.
Defense activities today represent 10 to 15% of the turnover of Mistral, which collaborates with the French, Singaporean and Luxembourg armies.
The company provides a software platform with AI agents, semi-autonomous assistants to which the user can delegate tasks, capable of aggregating a large number of data from heterogeneous sources.
“It’s very useful when you are in the general staff or when you have tactical coordination matters on the battlefield,” explained Mr. Mensch.
– Postures “d’idéologues” –
His group, which announced a five-year partnership with Airbus particularly in defense, also deploys small artificial intelligence models which can be put in weapons systems or defense systems to develop their autonomy, such as drones.
“Who are we to explain to soldiers, who know their job, who know the threats (…) what can be done?” he continued, refusing the postures of “ideologue”.
“The legitimacy of the Ministry of Defense is far greater than ours as a company that provides technology,” said Arthur Mensch.
“Our responsibility is obviously to choose who we work with” and to guarantee “absolute sovereignty to operate the weapons”, however continued the 33-year-old leader.
Its American competitor Anthropic made headlines after refusing to allow its AI tools to be used by the US Department of Defense for mass surveillance of citizens in the United States and to make weapons fully autonomous.
In retaliation, the Pentagon sued Anthropic and singled out all of its American rivals (SpaceX, OpenAI, Google, Nvidia, Reflection, Microsoft and Amazon subsidiary AWS) for its classified operations, including on the battlefield.
However, employees of Google’s artificial intelligence laboratory, Deepmind, subsequently mobilized in the United Kingdom against the military use of their technology by Israel and the American army, a sign of serious concerns on this subject.
– Security barriers –
Mistral also announced that it was now working on the development of artificial general intelligence (AGI), another explosive subject, particularly across the Atlantic, because it involves reaching the stage where AI would equal or even surpass all the intellectual capacities of humans.
“The objective is to have the most intelligent systems possible so that behind them, with the necessary adaptation, they have the greatest possible impact on businesses,” detailed Arthur Mensch.
Is there a risk of creating an out-of-control AI that would put humanity in danger?
“It makes absolutely no sense,” says the thirty-year-old. “We always deploy the models and then the systems in environments where we can unplug them.”
Security remains “a major concern” for his start-up, founded in 2023, which signed an agreement with Agence France-Presse (AFP) to use its news dispatches to respond to requests from its users.
This involves ensuring “that the model behaves as it has been told to do so” and imposing barriers on it which will “prevent it from taking unreasonable actions”, underlined Mr. Mensch. “This is an essential subject for our customers.”
On the cybersecurity side, Arthur Mensch confirmed that he is working on a specific product for businesses, at a time when European companies, particularly banks, have no access to the latest AI models such as Anthropic’s Mythos, described as extremely powerful for discovering computer vulnerabilities.
“Everyone needs to have cybersecurity systems that defend against attackers, who themselves are equipped with artificial intelligence,” he pointed out, adding that this product would be available this year.
