380,000 Vibe-Coded Apps Are Live on the Internet. 5,000 Are Leaking Your Medical Records.

Israeli cybersecurity startup RedAccess found 380,000 apps built with Lovable, Replit, Base44, and Netlify publicly accessible with virtually no security. About 5,000 of those are actively leaking medical records, financial data, customer chatbot logs, and corporate secrets. Default-public privacy settings. No auth. No rate limiting. Just exposed.

And it gets worse. Researchers found phishing sites impersonating Bank of America, FedEx, and McDonald's built on Lovable's platform. The same tool people use to prototype their startup is also the tool scammers use to build convincing phishing infrastructure. Fast.

This finding lands in the same week that Trend Micro reports MCP exposed servers have nearly tripled to 1,467, and OX Security demonstrated 94+ unpatched Chromium CVEs in Cursor and Windsurf. The pattern isn't subtle: the entire vibe-coding stack, from IDE to deployment platform, has security as an afterthought.

I'm not going to blame the tools. The tools are doing exactly what they promised: make it easy for anyone to build and ship apps. The problem is "ship" means "deploy to the public internet with default-open settings" and nobody in the vibe-coding workflow asks "should this endpoint require authentication?" The AI doesn't ask because it wasn't prompted to. The builder doesn't ask because they don't know to.

What should you do? Three things, today. First, audit the deployment settings on every app you've shipped through a vibe-coding platform. Check if your database is publicly accessible. Check if your API endpoints require auth. Second, if you're building with Lovable or similar tools, add "require authentication on all endpoints" and "make all data private by default" to your initial prompt. The AI will do it if you ask. Third, if you're running a team, treat vibe-coded prototypes like shadow IT. They're on your network. They're using your data. And right now, they're probably public.

The 5,000 apps leaking real data aren't hypothetical risk. That's someone's medical records, right now, indexed by search engines.