France's AI, So Hot Right Now

Here’s the sentence that rearranged the AI map this week. After a Trump directive pushed Anthropic to pull its newest models offline and set off a scramble for AI that doesn’t route through American servers, one name kept surfacing: Mistral.

The French company is suddenly everywhere, and mostly misunderstood. Its chatbot, Vibe (you might know it as Le Chat), gets pitched as “Europe’s ChatGPT,” which both sells it short and misses the point. Mistral’s real bet isn’t brand recognition, it’s independence: open models aimed at companies and governments that don’t want their data living inside a US tech giant. It’s running the Palantir playbook, parking engineers directly inside governments and big corporations to wire the AI into their systems.

And the money says it’s working. Mistral is reportedly raising around $3.5 billion at a $23.15 billion valuation, and its annual recurring revenue jumped from $20 million to over $400 million in a single year. That’s still a fraction of what US labs command, but it’s no longer a curiosity. It’s a real second option.

Which is the whole point. The moment America’s best model blinked offline, “who else can we use” stopped being a hypothetical. For anyone whose job touches a European client, a regulated industry, or a boss suddenly nervous about where the data goes, a credible non-American AI isn’t trivia. It’s leverage.

Two hundred fifty years after the Declaration of Independence, Google’s answer to “how do we sell Workspace” was to imagine Thomas Jefferson missing his deadline. The new commercial carries the tagline “Group project, but make it 1776.” Jefferson is mid-draft when Ben Franklin fires off a nagging text. Edits pile up in Google Docs. Someone schedules a meeting in Google Calendar. Democracy, apparently, needed a shared calendar.

It’s a cute ad. It’s also a tell. When a trillion-dollar company markets AI by showing it doing group-project logistics, that’s the version of AI they actually expect to sell: the boring, useful, sitting-inside-the-doc-you-already-use kind.

On June 29, an anonymous account posted a tool for Archive of Our Own that hunts for a fingerprint Claude leaves in copy-pasted text, a hidden code tag called “font-claude-response-body.” Land on a story written that way and the whole page glows red. The Verge tested it. Paste straight from Claude, red screen. Paste the exact same story through anything in between, nothing. The signal is loud and the loophole is enormous.

That gap is the whole story. Communities are already using the red flag to publicly name and shame flagged writers, and the tool only catches the laziest possible method while missing anyone who copies through a notes app first. Every AI detector has this problem. It punishes the careless and clears the crafty, then everyone treats the output as proof.

Security researchers just walked through the 2026 Vercel incident, and the origin story is uncomfortably ordinary. No zero-day, no nation-state. An employee installed an AI tool that hadn’t been vetted, it carried Lumma Stealer malware, and that malware quietly scooped up corporate credentials, active browser session cookies, and OAuth tokens sitting on the machine. Attackers used those to walk into internal systems, dig through customer environment variables, and demand a $2 million ransom.

This is what “shadow AI” means in practice. Every unapproved extension, every browser plugin you grant permissions to with your work login, is a door. Most are harmless. You only need to be wrong once.

Michelle Turner had no MBA, no startup pedigree, and no venture backers lining up for a first-time solo female founder. What she had was a problem she knew cold, foster children can’t get mental-health care fast enough, and a set of AI tools she treated like a graduate school. Working from her home in Virginia Beach, she used AI to teach herself startup culture, write a business plan, and sharpen an investor pitch until the funding actually came.

Here Now Health launched in January 2025. It now has 16 employees and is certified in three states to provide Medicaid-funded counseling for kids entering the foster system. “It was like going to a master’s level class every day from the robot,” Turner told Reuters. “It was my startup advisor.” Note what this isn’t: it isn’t an AI company. It’s a mental-health company that used AI to skip the part where you need money and pedigree to get started.

That’s the shift worth clocking. The stuff that used to gate ambition, the expensive advice, the polished deck, the insider fluency, just got a lot cheaper to access. The advantage now goes to the person who knows exactly what they’re building and uses AI to close the gap.