France Taps Mistral for AI Security Tests After 700K Taxpayer Breach; Bitcoin (BTC) in Focus

Crypto News

France will use sovereign AI providers such as Mistral, rather than OpenAI, to test government systems for security vulnerabilities, Budget Minister David Amiel said Tuesday after a cabinet meeting in Paris. The decision follows a breach at the national tax agency that exposed data of roughly 700,000 taxpayers. “This excludes OpenAI,” Amiel told reporters, adding that the state would rely on what he called sovereign AI companies. The government plans to use AI tools to probe its own services for weaknesses. The tax-office intrusion was disclosed the previous Thursday. Tax agency chief Amélie Verdier announced a second data breach on Monday, saying the incident was still being assessed. The first breach involved an attacker using a compromised internal VPN and a stolen identity to reach a taxpayer search tool; access was traced to late June 2026 and closed by the end of that month. The finance ministry said the stolen data includes names, birth details, postal and email addresses, phone numbers, tax identification numbers, withholding rates and correspondence with officials. Usernames and passwords were not among the leaked records. France’s data-protection authority, CNIL, has been notified, and the DGFiP said it will file a criminal complaint. Mistral, a Paris-based startup backed by chip-equipment maker ASML, has become France’s preferred provider when it wants to show it can run critical systems without depending on US technology. In June, Amiel unveiled a government plan called “Notre IA” to distribute sovereign AI tools across public services. The state digital agency DINUM has built an assistant called “L’Assistant” on a Mistral model, hosted in SecNumCloud-certified data centers, and is rolling it out to nearly one million government employees. The planned use of automated scanners resembles an AI trading bot that executes under rules set by a human operator. The security-testing role extends that strategy, keeping France’s AI infrastructure under national control.

Mozilla is testing an opt-in “Smart Window” in Firefox, a separate browser window that lets users chat with an AI assistant of their choice while keeping the feature switched off by default. The company opened a waitlist at firefox.com/ai for early testers, describing the project as “user-controlled” in its own announcement. The window sits alongside Firefox’s Classic and Private modes, and users can pick the model, decide when to use it, or disable it entirely. Mozilla says the design is meant to avoid locking anyone into a single AI ecosystem, drawing a contrast with browsers built by AI companies that present a binary choice: use AI all the time or not at all. The feature extends existing Firefox tools such as the sidebar chatbot on desktop and Shake to Summarize on iOS. It also builds on the Project Nova redesign, which introduced a single control to turn off every AI feature at once. A one-click AI kill switch arrived earlier this year. For crypto users, this opt-in architecture resembles the permission model of an AI crypto wallet, where the user, not the provider, decides when an automated service is allowed to act. Mozilla is inviting feedback from the first testers and says AI should be “open, accessible, and driven by choice.” The waitlist remains open, and the feature stays off unless a user turns it on. The broader browser market is split between privacy-first assistants and AI-centric rivals: Brave has pushed its LEO assistant, while Perplexity Comet, Microsoft Edge and Opera Neon have made AI central to the browsing experience. OpenAI abandoned its own Atlas browser to focus on a broader agentic app, and Google experimented with a locally downloaded model. Mozilla’s bet is that the off-switch, not the assistant itself, is the differentiator, and the open-source code behind these tools is what lets outside researchers verify their claims.

Both stories turn on who controls the AI layer and whether that control can be revoked. Mozilla is selling an off-switch; France is building a sovereign alternative to American AI. For Bitcoin (BTC), trading near $64,000 at press time, the principle is familiar — a neutral, permissionless network is most resilient when no single party can force participation. The primary-source record supports the shift: Mozilla’s own announcement frames the feature as user-controlled, and Amiel’s statement after the cabinet meeting explicitly excludes OpenAI. The same self-custody ethos is now shaping browser design and state security policy, making the altcoin governance debate more relevant. Algorithmic stablecoins rely on code-based rules that must be auditable; these AI systems depend on open, revocable controls.