RedAccess Finds More Than 5,000 Vibe-Coded Apps on Lovable, Replit, Base44 and Netlify Running With No Authentication

Overview

A scan published this week by the Israeli cybersecurity firm RedAccess found more than 5,000 applications built on AI vibe-coding platforms running with virtually no security or authentication, with roughly 40 percent of those apps exposing sensitive corporate and personal data to anyone on the open web, according to Slashdot. The platforms named in the report are Lovable, Replit, Base44 and Netlify.

RedAccess researcher Dor Zvi told reporters that “organizations are actually leaking private data through vibe-coding applications” and called the scope “one of the biggest events ever where people are exposing corporate or other sensitive information to anyone in the world,” as quoted by Slashdot.

What We Know

The broader scan, first reported by Axios on May 7, covered roughly 380,000 publicly accessible assets built with tools from Lovable, Base44, Replit and Netlify; about 5,000 of those held sensitive corporate data, a figure also reported by VentureBeat. Slashdot’s account of the report describes the same four platforms and the same 5,000-plus figure for apps with no authentication, with roughly 2,000 of those revealing private information on inspection, according to Slashdot.

The data RedAccess pulled out of the exposed apps spans hospital work assignments containing physician personally identifiable information, retail chatbot conversation logs with customer names and contact details, shipping firm cargo records, and corporate go-to-market presentations, according to Slashdot. Axios reported additional examples it independently verified, including an app for a shipping company that detailed which vessels were going to which ports, an internal application for a health company that detailed active clinical trials across the United Kingdom, full unredacted customer service conversations for a U.K. cabinet supplier, and internal financial information for a Brazilian bank.

The root cause cited in the report is a default-public configuration: privacy settings on several of the vibe-coding tools left applications publicly accessible unless the user manually toggled them to private, and many of those apps were then indexed by Google and other search engines, according to Axios. On Lovable’s domain specifically, RedAccess also identified phishing sites impersonating Bank of America, Costco, FedEx, Trader Joe’s and McDonald’s, according to Slashdot.

Platform Responses

Replit chief executive Amjad Masad said RedAccess gave the company only 24 hours of notice before going to the press, and posted on X that Replit lets users choose whether their apps are public or private and that public-by-choice apps being reachable on the internet is the expected behavior, with privacy settings changeable at any time, according to Axios.

Spokespeople with both Base44 and Lovable told Axios that RedAccess’s reports to the companies did not include the URLs that would have allowed them to investigate or verify the findings. By Wednesday, some of the exposed applications Axios had viewed earlier in the week were no longer publicly accessible after the platforms were notified.

What We Don’t Know

RedAccess has not publicly released the underlying URL list, and the platforms named in the report dispute the methodology, so an independent count of how many of the 5,000-plus flagged apps had been actively accessed by malicious parties before remediation is not available, per the Axios and VentureBeat accounts. The split between misconfigured user accounts and platform-level defaults also varies by tool: each of the four named platforms exposes its own privacy controls differently, and the report does not break the 5,000 figure down by vendor in the public summaries.

It is also unclear what fraction of the 380,000 baseline scan represents abandoned demos and prototypes versus production-grade internal tools. Zvi framed the issue as a shadow-IT problem in which “anyone from your company at any moment can generate an app, and this is not going through any development cycle or any security check,” according to Slashdot, but the report does not quantify how many of the exposing organizations had any policy on AI-built internal applications.