- Mistral CEO Arthur Mensch told a French commission of inquiry that the French army’s code bases should not be scanned by Anthropic’s Mythos model.
- Mensch said the resulting dependency would be ‘nearly impossible to reverse’.
- Modern models, per Mensch, can now orchestrate attacks, detect vulnerabilities, and suggest exploits — a capability Mythos has demonstrated.
- Mistral’s US investors hold less than 30% of the company; a sale is not planned and the company aims to go public eventually.
What Happened
Mistral AI co-founder and CEO Arthur Mensch warned a French commission of inquiry about Europe’s growing cybersecurity dependency on US-built frontier AI models, The Decoder reported on Friday. Referring specifically to Anthropic’s Mythos model, Mensch said modern frontier models can orchestrate attacks, detect vulnerabilities, and suggest exploits. He argued the French army’s code bases should not be scanned by Mythos, as it would create a dependency that is “nearly impossible to reverse.”
Why It Matters
Mensch’s testimony makes explicit a debate that has been escalating across European policy circles through 2026: whether sovereign AI capability requires native European models for critical-infrastructure applications, even when US-built models are more capable. The argument has two layers. First, the strategic-autonomy layer — once a code base is scanned by Mythos, the model’s training and architecture become part of the operational dependency stack. Second, the technical-capability layer — the same capability that lets Mythos defend code can be repurposed to attack it.
The framing is also commercially relevant for Mistral itself. The company has positioned itself as the only EU-based provider with competitive frontier-tier language models. A French government decision to source from Mistral rather than Anthropic would extend that positioning to defence-grade applications.
Technical Details
Mensch acknowledged that the same vulnerabilities Anthropic’s Mythos could detect could also be found by Mistral’s own models or by Chinese models. The strategic concern, per Mensch, is the dependency structure rather than the technical capability per se. The EU is currently negotiating with both OpenAI and Anthropic for early access to their most capable cybersecurity models — a programme that runs in parallel to the sovereignty debate Mensch is amplifying.
Mensch also addressed Mistral’s ownership: US investors hold less than 30% of the company. European capital was preferred but not available in sufficient quantity, leading to the current cap-table structure. Mensch said Mistral has no current sale plans and intends to remain independent and eventually pursue a public listing.
Who’s Affected
The French government and EU policymakers face an immediate decision on procurement of frontier-AI cyber capabilities. Anthropic faces a public counter-narrative on its European enterprise expansion. Mistral gains both regulatory and commercial momentum: its public positioning aligns directly with European sovereign-AI procurement priorities. Other EU AI players — Aleph Alpha (now part of Cohere), Stability AI’s European entities, Black Forest Labs — face the question of how to align with the sovereignty narrative. US frontier-lab competitors — OpenAI, Google DeepMind, xAI — watch the precedent that France’s procurement decisions will set for the wider European market.
What’s Next
The French commission of inquiry will publish its findings on a timeline not yet disclosed. EU-level negotiations with OpenAI and Anthropic on cyber-capability access continue in parallel. Mistral’s IPO timing is not announced; the company has indicated public-market exit as its planned trajectory but has not committed to a specific year.
