Lovable Security Breach: What the April 2026 Exposure Means

What happened

On April 20, 2026, a security researcher publicly showed that Lovable, the AI "vibe coding" platform, let users see data in projects that belonged to other people. Lovable says it shipped a fix within two hours. It also admits that both the product and its first public response "missed the mark." 1 The company's account is narrow. It says chat history and source code in public projects could be reached by any logged-in Lovable user who had a project link, between February 3 and April 20, 2026. It says private projects and Lovable Cloud were never affected. 1

Lovable traces the cause to a February 2026 backend regression. That change reopened access to chat histories on public projects and undid protections the company had added between March and November 2025. 1

Where the accounts diverge

Outside accounts describe a wider problem than Lovable's post does. The Next Web calls the flaw a broken object-level authorization (BOLA) bug in Lovable's API. It reports that anyone with a free account could pull another user's profile and project data. 2 It also says the hole stayed open for 48 days after Lovable closed a bug bounty report on it without escalating. 2 That detail does not appear in Lovable's own write-up.

Security firm Bastion goes further on scope. It says the exposure covered every Lovable project created before November 2025, not just public ones. By its account, the leaked data included source code, Supabase database credentials, AI chat transcripts, and customer data stored in connected backends. 3 Bastion credits the disclosure to a researcher posting as @weezerOSINT. It describes the root cause as a tenant isolation failure layered on top of the missing Row Level Security pattern already flagged in CVE-2025-48757. 3

The two framings can partly be reconciled. Lovable's post focuses on the regression window and on public projects. The outside reports point to older weaknesses that predate February, especially apps whose databases lacked proper access rules. Readers should treat the exact scope as disputed. On one point all sides agree: authenticated strangers could reach data they should not have seen.

A pattern, not a one-off

The Next Web counts three documented security incidents at Lovable over roughly two months. Together they exposed source code, credentials, AI chat logs, and personal data from thousands of users of apps built on the platform. 2 Bastion's link back to CVE-2025-48757 supports the same picture: these are recurring weaknesses, not a single slip. 3

The wider industry numbers are not reassuring either. The Next Web cites research suggesting up to 62% of AI-generated code contains vulnerabilities. 2 A separate comparison of AI app builders cites a Stanford finding that 80% of AI-built apps ship exploitable code. 5

Why it stings for Lovable specifically

The timing collides with Lovable's main selling point. Comparisons of the leading AI builders often call Lovable the best option for non-technical founders. The reason is that it bundles Supabase-based authentication, a database, and hosting for about $25 a month. 5 Rivals are positioned differently:

• Bolt.new targets developers, offering a real in-browser Node.js environment and a generous free token allowance. 5
• Vercel's v0 produces polished React front ends but still relies on outside tools for backend, auth, and database logic. 45

That gap is exactly what Lovable fills. Its customers are often people who cannot audit access-control rules themselves, which is why they chose a tool that handles auth for them. When the platform's own tenant isolation fails, those users have the fewest ways to notice or respond.

The stakes are high on the business side. Lovable reportedly passed $200 million in annual recurring revenue. It raised a $330 million Series B at a $6.6 billion valuation, with backers including NVIDIA, Salesforce, Databricks, and Atlassian. 5 The Next Web puts its user base at around eight million. 2

The verdict

The two-hour fix shows Lovable can move quickly once a problem is public. That speed matters less than the reported failure to escalate an earlier bug report, and less than a regression that quietly removed protections the company had already built. 12

For founders who picked Lovable because auth seemed "handled," the practical lesson is to verify rather than trust:

• Rotate any Supabase credentials.
• Confirm Row Level Security is enabled on every table.
• Assume older projects may have leaked data.

Bastion's guidance points the same way. 3

More broadly, this episode suggests that "easiest" and "secure by default" are separate promises. Vibe coding platforms have shown they can deliver the first. They have not yet proven the second.