A security researcher has exposed a glaring vulnerability in AI app-building platform Lovable that allowed any free account to access other users' project source code, database credentials, and AI chat histories. The bug, a classic object-level authorization failure (BOLA), was reported via HackerOne on March 3, 2026, and remained unpatched for 48 days.
https://twitter.com/weezerOSINT/status/2046170666131669027
In a demonstration, researcher @weezerOSINT accessed a project belonging to the Danish nonprofit Connected Women in AI. He pulled the entire source code for its admin backend and read conversations where developers discussed database table structures with Lovable's AI—including sensitive fields like email, first_name, and last_name.
His testing revealed a patchy fix: new projects created in April 2026 returned a 403 Forbidden error, but older projects the same developer was editing just 10 days earlier returned a 200 OK with a full source file tree. Lovable had apparently applied permission checks only to new projects, leaving existing ones exposed.
The company's response was a rollercoaster. Initially, Lovable dismissed the issue as "intentional design" and blamed unclear documentation. Later, it admitted a mistake, explaining that a February 2026 backend permission overhaul accidentally reopened chat access for public projects. Lovable then pointed the finger at HackerOne's triage team, claiming they closed the report because they believed "chat for public projects is viewable" was expected behavior.
Lovable, which boasts a $6.6 billion valuation and clients like Uber, Zendesk, and Deutsche Telekom, now joins a growing list of AI platforms grappling with basic security oversights. This incident echoes broader industry concerns, such as when Meta plans to cut 8,000 jobs in May as AI reshapes its workforce, highlighting the rapid, sometimes reckless, scaling in the sector. Meanwhile, as companies like Anthropic's Claude Opus 4.7 arrives with smarter coding, better vision, and tighter safety, it's clear that security can't be an afterthought in the race for AI dominance.