BNP Paribas and Mistral AI announced on May 26 in Paris a cybersecurity shift in their partnership.The first eurozone bank turns to France’s AI champion to build a sovereign alternative to Mythos, Anthropic’s model, which no European bank can access.

Anthropic is acting as the gatekeeper, and European banks are not allowed in

Mythos, the cybersecurity AI model that Anthropic launched on April 7 through Project Glasswing, remains restricted to around forty handpicked organizations. The list includes JPMorgan Chase, AWS, Apple, Google, Microsoft, and CrowdStrike, but not a single European bank. The European Commission has been negotiating access since April. According to Spanish officials quoted by the press, discussions are “at a standstill,” and the Bundesbank had to formally intervene to support Brussels.

The problem is concrete. Mythos can identify critical software vulnerabilities and produce functional exploits in more than 70% of cases during testing. In six weeks, it has detected more than 10,000 vulnerabilities in open-source software used by global financial infrastructure. Frank Elderson, vice-chair of the ECB Governing Council, has warned eurozone banks that without a comparable tool, defenders will be “structurally behind” attackers.

The French “Plan B” is not one: it has existed since 2023

The cybersecurity pivot announced on May 26 is not a sudden decision. BNP Paribas has been working with Mistral AI since September 2023, when its Global Markets division began testing its models.The collaboration was extended across the entire group in February 2024, then formalized through a multi-year agreement signed in July 2024. Today, BNP uses Mistral models for its virtual assistants in retail banking, document extraction, equity research and regulatory compliance.

Marc Camus, Chief Information Officer of BNP Paribas, reframed the debate at the press conference: “There is a lot of noise in the market about Mythos and whether Mythos is available to certain banks, particularly European banks. But let’s not forget that there are other models from other companies.” To put it plainly: BNP has no intention of waiting for Anthropic to deign to open its doors.

On Mistral’s side, Corentin Petit, worldwide head of solutions, confirmed that the startup was developing a model dedicated to cybersecurity and was discussing with several European banks in addition to BNP. The company has not provided a release date or publicly demonstrated that its model could compete with Mythos. But the bet European banks are making is clear: a sovereign model, even if slightly less performant, is better than no model at all.

Mistral indeed has the resources to match its ambitions. In April, the startup secured an €830 million loan from seven banks (including BNP Paribas and Crédit Agricole CIB) to finance a data center in Bruyères-le-Châtel, Essonne, equipped with 13,800 Nvidia Grace-Blackwell chips and set to become operational as early as the second quarter of 2026.The Ministry of the Armed Forces signed a framework agreement with Mistral in December 2025. For French BNP Paribas customers, the promise is simple: their bank will not rely on an American company banned by its own government to protect their data.

👉🏻 Follow the latest tech news in real time: add 01net to your sources on Google and subscribe to our WhatsApp channel.

Source: Bloomberg