Lovable Exposed Every User's Source Code and Database Passwords

Lines of code displayed on a computer monitor in a dark room — Photo via Unsplash

TL;DR: Lovable, a “vibe coding” platform valued at $6.6 billion with 8 million users, had a Broken Object Level Authorization (BOLA) flaw that let any free account access other users’ source code, Supabase database credentials, Stripe customer IDs, and full AI conversation histories. A security researcher reported the bug on March 3, 2026. Lovable ignored it for 48 days, then went public denial → blame HackerOne → blame documentation → partial apology. Real-world impact: 18,697 student records exposed from a university app, including 4,538 minors. This is the third major Lovable security incident in 13 months. If you built anything on Lovable before November 2025, your credentials are likely compromised.

Five API Calls. That’s All It Took.

On April 20, 2026, security researcher @weezerOSINT posted a warning on X that set the AI coding world on fire: “Lovable has a mass data breach affecting every project created before November 2025” ^[1].

The vulnerability was embarrassingly simple. Lovable’s /projects/{id}/* API endpoints verified that a user had a valid Firebase authentication token, but never checked whether that user actually owned the project they were requesting. Five API calls from a free account. That’s the entire exploit ^[2] ^[3].

As one researcher put it: “This is not hacking. This is five API calls from a free account” ^[3].

Here’s what anyone could pull from any Lovable project created before November 2025:

Full source code: every line of every project
Hardcoded Supabase credentials: database connection strings and service keys
AI chat histories: every prompt, every response, every piece of data pasted into the conversation
Personal data: names, job titles, LinkedIn profiles, Stripe customer IDs
Customer data: whatever end-user data lived in connected backends

The technical name for this bug is BOLA, Broken Object Level Authorization. It’s #1 on the OWASP API Security Top 10. Not obscure. Not novel. The kind of flaw that would fail a junior developer’s code review ^[2].

18,697 Students. 4,538 Minors. All Exposed.

This wasn’t theoretical. Researchers demonstrated the impact by accessing a Connected Women in AI nonprofit’s admin panel. Five API calls pulled database records containing employee information from Accenture Denmark and Copenhagen Business School ^[3].

In February 2026, security researcher Taimur Khan found something worse. A featured university-affiliated app built on Lovable had exposed 18,697 student records, including 4,538 minors from UC Berkeley and UC Davis. Full PII. Khan submitted a support ticket to Lovable. It was closed without a response ^[4] ^[5].

Khan also found 16 separate vulnerabilities in that single Lovable-hosted app. Six were rated critical ^[5].

Organizations whose data was accessible through the Lovable flaw included employees at Nvidia, Microsoft, Uber, and Spotify ^[2]. Uber, Zendesk, and Deutsche Telekom use the platform directly ^[1].

48 Days of Denial, Deflection, and Blame

The researcher reported the BOLA vulnerability to Lovable’s HackerOne bug bounty program on March 3, 2026. What followed was a masterclass in not taking responsibility ^[1] ^[2] ^[6].

March 3: Bug reported to HackerOne.

48 days of silence. HackerOne labeled the report a “duplicate submission” and left it open. No fix. No acknowledgment ^[6].

April 20: @weezerOSINT goes public. Lovable responds with Statement #1: “We did not suffer a data breach.” They called the exposed data “intentional behaviour” related to public project visibility settings ^[1].

Statement #2: After backlash, Lovable acknowledged that early free-tier users didn’t even have the option to make projects private. But still: not a breach. Just “unclear documentation” about what “public” means ^[1] ^[2].

Statement #3: Lovable admitted that a February 2026 backend permissions change accidentally re-enabled access to chat histories on public projects. Then blamed HackerOne: “Reports were closed without escalation because our HackerOne partners thought that seeing public projects’ chats was the intended behaviour” ^[1] ^[2].

Cybernews summarized it best: the company went on an “ego trip denying vulnerability, then blames others for said vulnerability” ^[2].

HackerOne’s response? “Given the nature of customer programs and the need to review details carefully, we’re not able to comment further right now” ^[1].

Third Major Incident in 13 Months

The April 2026 disclosure isn’t Lovable’s first rodeo. It’s their third major security event since March 2025 ^[4] ^[5]:

March 2025: Matt Palmer (Replit) privately disclosed a Row Level Security misconfiguration in Lovable-generated Supabase apps
April 2025: CVE-2025-48757, with a CVSS score of 8.26–9.3. Out of 1,645 scanned Lovable apps, 170 were vulnerable, with 303 exposed endpoints
April 2026: The BOLA flaw exposing every pre-November 2025 project

Three incidents in 13 months. Each one a fundamental failure of access control, the most basic security requirement for a multi-tenant platform. And this is the company that just raised funding at a $6.6 billion valuation ^[5].

The Vibe Coding Security Problem

Lovable isn’t just a company with bad security. It’s a symptom of a much bigger problem.

“Vibe coding” (using AI to generate entire applications from natural language prompts) exploded in 2025. Lovable, Replit, Bolt, and others let anyone build a full-stack app without writing a single line of code. The pitch: democratize software development. The reality: millions of apps built by people who don’t know what Row Level Security is, running on platforms that apparently don’t either ^[5].

The numbers are ugly ^[5]:

40–62% of AI-generated code contains security vulnerabilities
91.5% of vibe-coded apps had at least one AI hallucination-related vulnerability in Q1 2026
2.74x higher flaw rate in AI-written code vs. human-written code
60%+ of vibe-coded apps exposed API keys or database credentials
70% of Lovable apps had Row Level Security disabled entirely

And yet: 87% of Fortune 500 companies have adopted at least one vibe coding platform. Enterprise adoption grew 340% year-over-year. Non-technical user adoption surged 520% ^[5].

More people building apps that handle sensitive data. Fewer of them understanding what they’re building. And the platforms they’re building on can’t even secure their own APIs.

A Pattern: AI Tools as Attack Surface

The Lovable breach landed two days after Vercel confirmed its own breach, caused by an AI evaluation tool called Context.ai that gave attackers a path into Vercel’s infrastructure. And days after OX Security disclosed a design-level vulnerability in Anthropic’s Model Context Protocol affecting 200,000+ AI servers.

Three AI platform security failures in a single week. The pattern is clear: the tools we’re adopting to move faster are creating attack surfaces we don’t understand. And the companies building them are prioritizing growth over the basics.

If You Used Lovable: Do This Now

Within 24 hours:

Rotate every credential. All Supabase keys, API tokens, and third-party service credentials used in Lovable projects. Assume they’re compromised
Enable Row Level Security. Check every Supabase table. If RLS is disabled, enable it and write proper policies
Pull Supabase audit logs. Look at the last 60 days of database access for reads you don’t recognize
Check your AI chat histories. Did you paste API keys, passwords, or sensitive data into Lovable’s chat? Rotate those credentials too

Within 72 hours:

Export and archive project data. Don’t leave credentials sitting in Lovable’s systems
Audit connected SaaS accounts. If Lovable had access to your Stripe, check for unauthorized activity
Notify affected users. If your Lovable app handled customer data, you may have GDPR or state privacy law notification obligations
Consider migrating. Three major incidents in 13 months is a trend, not a fluke

References

Published: April 23, 2026