Lovable Left the Door Open for 48 Days and Called It a Feature

A security researcher found a flaw in Lovable’s API that let anyone with a free account access other users’ source code, database credentials, and AI chat histories. Lovable’s response? Close the bug report, deny it was a breach, blame HackerOne, and leave existing projects exposed for 48 days.

This is a $6.6 billion company with eight million users. Nvidia and Microsoft teams use this platform. And the vulnerability was a textbook BOLA — Broken Object Level Authorization — the kind of bug that makes it into every “intro to API security” course ever written.

Five API Calls to Your Secrets

Here’s how simple this was. With nothing more than a free Lovable account, an attacker could hit the API and pull another user’s profile, their public projects, the full source code for those projects, and — here’s the fun part — hardcoded Supabase database credentials embedded in that source code.

Five API calls. No authentication bypass needed. No exotic exploit chain. The API just… handed it over. Like a valet parking attendant who gives the keys to whoever asks nicely.

Every project created before November 2025 was affected. Tens of thousands of developers. Their end users. Their databases. All accessible to anyone who could type a curl command.

The Response Was Worse Than the Bug

The researcher reported the flaw to Lovable’s bug bounty program on March 3. Lovable patched it for new projects but never fixed it for existing ones. When the researcher filed a follow-up report pointing out that older projects were still exposed, Lovable marked it as a duplicate and closed it.

Then came the public statements. Lovable initially denied it had suffered a breach. Then they claimed the exposure was “intentional” — as in, they meant to make that data accessible. Then they shifted blame to HackerOne. Only after the story went public and the backlash hit did they acknowledge their mistake and revoke access to chat data on public projects.

That’s four separate responses, each worse than the last. Deny, deflect, blame the platform, then quietly fix it when people are watching. A masterclass in how not to handle a security incident.

The Vibe Coding Security Stack

This isn’t just a Lovable problem. It’s a structural one. Vibe coding platforms are optimized for one thing: getting you from prompt to deployed app as fast as possible. Security is an afterthought bolted on after the demo video is shot.

The same week Lovable was leaving doors open, Vercel got breached through a compromised AI tool, and the Bitwarden CLI was hijacked in a supply chain attack where the malware specifically hunted for credentials to Claude, Cursor, and Codex. Three platforms, one week, all connected to the AI-assisted development ecosystem.

The attack surface isn’t just your code anymore. It’s every platform that touches your code, every tool in your pipeline, and every AI assistant that has write access to your repo.

What This Means for You

If you’ve ever used Lovable, assume your source code was accessible. If that source code contained any credentials — API keys, database URLs, auth tokens — rotate them now. Not tomorrow.

And if you’re building on any vibe coding platform, start asking the uncomfortable questions: What happens to my source code on their servers? Who can access it? What’s their bug bounty response time? Because right now, the answer at a $6.6 billion platform was “48 days and a shrug.”

The tools are getting better at writing code. They are not getting better at keeping it safe.