Lovable Breach: 8M Users, 48 Days Exposed

8 million users. $6.6 billion valuation. 48 days of exposed database credentials, source code, and customer data. Lovable, the vibe coding platform backed by Accel and EQT, just became the largest security failure in the AI coding tool space.

On April 20, a security researcher operating as @weezerOSINT on X posted findings that shook the developer community: anyone with a free Lovable account could access other users' projects, read their AI chat histories, extract their Supabase database credentials, and download their source code. The entire attack required five API calls and zero technical skill.

The worst part? The researcher reported this vulnerability through HackerOne on March 3, 2026. Lovable's bug bounty partners closed the report without escalation because they believed the exposed data was "intentional behavior."

What was exposed

The vulnerability was a Broken Object Level Authorization (BOLA) flaw, ranked #1 on the OWASP API Security Top 10. Lovable's API endpoints verified that a user was logged in but never checked whether that user owned the project they were requesting. That single missing check opened the entire platform's project history to anyone with a free account.

The exposed data included:

• Full source code for projects created before November 2025
• Hardcoded Supabase database credentials embedded in that code
• Complete AI chat histories (every prompt, every response, every error log a developer shared with the AI)
• Customer data from live production databases accessible through the leaked Supabase keys

One confirmed example: Connected Women in AI, a Danish nonprofit, had its project data exposed. The researcher extracted real names, job titles, LinkedIn profiles, and Stripe customer IDs belonging to professionals at Accenture Denmark and Copenhagen Business School.

Employees at Nvidia, Microsoft, Uber, and Spotify reportedly have Lovable accounts tied to affected projects.

Lovable's response made it worse

Lovable's public communications cycled through denial, deflection, and a partial apology within a single day.

The company posted on X that it "did not suffer a data breach" and attributed the exposed data to "intentional behavior," blaming its own documentation for being "unclear." Hours later, the company threw its bug bounty partner HackerOne under the bus, claiming reports were "closed without escalation" because reviewers thought the behavior was by design.

By the end of the day, Lovable admitted that a February backend update had "accidentally re-enabled access to chats on public projects."

The company applied a fix for newly created projects. Projects created before November 2025 (the ones with the most accumulated data, the most credentials, and the most customer information) remained exposed.

Vibe coding has a structural security problem

Lovable isn't an outlier. It's representative of a pattern playing out across the entire vibe coding category, including platforms like Bolt.new and Replit Agent.

Research from Q1 2026 paints a grim picture:

• Between 40% and 62% of AI-generated code contains security vulnerabilities, depending on the study
• AI-written code produces flaws at 2.74 times the rate of human-written code
• An assessment of over 200 vibe-coded applications found that 91.5% contained at least one vulnerability traceable to AI hallucination

The pattern is consistent. AI coding tools generate full-stack applications in minutes. Those applications connect to real databases, real API keys, and real customer data. The developers using these tools often lack the security expertise to audit what the AI produced. The platforms themselves prioritize growth over security hardening.

I wrote about this pattern in my vibe coding security checklist earlier this year. The Lovable breach validates every concern on that list.

What you should do right now

If you ever built anything on Lovable, especially before November 2025, these steps are not optional. Treat this as an active incident response.

Step 1: Rotate every credential that touched Lovable

This means Supabase keys, API tokens, database passwords, and any third-party service credentials you connected to a Lovable project. Don't wait for Lovable to confirm whether your specific project was accessed. The BOLA flaw was exploitable for 48 days. Assume your credentials are already in someone else's hands and act accordingly.

Step 2: Audit your Supabase Row Level Security

Lovable-generated code frequently ships without proper RLS policies. If your Supabase database has RLS disabled or misconfigured, your customer data is accessible to anyone who obtained the service role key from the breach. Check every table. Enable RLS on every table. Test it with a non-admin user to confirm it actually works.

Step 3: Review your AI chat history for leaked secrets

Developers routinely paste error logs, database schemas, and credentials into AI chat interfaces during debugging sessions. If you did this on Lovable, those conversations were readable by any free-tier user for months. Every secret you shared in a Lovable chat should be considered compromised and rotated.

Step 4: Stop storing production credentials in AI coding tools

This should be a written policy for any team shipping code. Use environment variables, secrets managers (AWS Secrets Manager, HashiCorp Vault, Doppler), or dedicated vault services. Never paste a production key into a chat prompt, regardless of the platform.

Step 5: Add vibe coding platforms to your vendor risk assessment

If your organization builds on Lovable, Bolt.new, Replit Agent, or any other AI app builder, those platforms need the same security scrutiny as any other data processor. Include them in quarterly tabletop exercises. Require penetration testing on anything that touches production data. The Lovable breach is the third documented security event on that platform in just over a year.

Credentials are still the #1 attack vector

This breach didn't require a zero-day exploit, AI-powered hacking tools, or nation-state resources. It required five API calls from a free account. The attacker walked through the front door because the door was never locked.

I keep coming back to the same finding from my breached password analysis: 73% of compromised credentials follow predictable patterns, and most breaches start with a single exposed password or key.

The Lovable breach is a credential exposure problem at platform scale. Every Supabase key, every API token, every database password that Lovable's BOLA flaw made accessible is now a potential entry point for credential stuffing, lateral movement, and data exfiltration.

If you're building with any AI coding tool that generates full-stack applications from prompts, the vibe coding security checklist covers the minimum steps to avoid ending up in the next breach headline.

Your credentials are the keys to everything. Treat them that way.