French AI flagship Mistral AI finds itself at the heart of a major cyber turmoil. A malicious actor, operating under the pseudonym "mrwho," claims to have exfiltrated the company's entire source code. This announcement, posted on a cybercrime forum, suggests a deep compromise of the Parisian startup's infrastructure. If confirmed, this would mark the second breach in just a few months, following a suspected incident last May. For now, the company's management has not officially confirmed the veracity of the leak, but the scope of the disclosed data raises legitimate concerns regarding the security of this champion of European digital sovereignty.

Analysis of the samples shared by the hacker reveals a list of 339 files touching the technological core of Mistral AI. The archives reportedly contain confidential projects intended exclusively for internal use, including inference tools, fine-tuning scripts, and private development libraries. Beyond the models themselves, the company's logistical inner workings appear to be exposed: from GPU management dashboards and Kubernetes infrastructure configurations to modules related to billing and compliance processes. While no active login credentials or API keys have been detected in the initial extracts analyzed, the structural nature of this data exposes the internal logic and proprietary architecture of the platform.

Unlike many cyberattacks targeting the general public, this leak does not appear, for the moment, to involve personal user data. The damage is almost exclusively concentrated on intellectual property, the most valuable asset of any cutting-edge tech firm. The disclosure of prototypes, such as a web agent integrated into the browser, highlights the strategic edge that the hacker is attempting to monetize. The question remains, however, whether this September attack is a genuine new intrusion or simply a recirculation of data stolen earlier in the year. The presence of recent code modifications within the files would be the key technical indicator to validate the persistence of a security vulnerability within the company's systems.

This case illustrates a fundamental paradigm shift in the artificial intelligence ecosystem: computer code is paradoxically becoming less critical than training data and internal workflows. In an era where AI models can themselves help reconstruct complex software in record time, value is shifting toward business processes and the expertise accumulated over time. This incident reinforces the growing caution among major industry players, who are now imposing drastic security protocols to protect their industrial secrets. For Mistral AI, the challenge is now to secure its technical assets while preserving the trust of its institutional partners, as the protection of proprietary data emerges as the new essential strategic bulwark.