5,000 Vibe-Coded Apps Just Leaked Corporate Data. Finance Was in the Blast Radius.
Last week, I wrote about how vibe coding is a genuine productivity tool for finance teams … right up to the point where it touches production data or anything an auditor might ask about. And I stressed the importance of having proper security measures in place to protect proprietary data before moving into production.
This week, those warnings got a body count.
Security researchers at Israeli cybersecurity firm RedAccess published findings on May 7 that quantify what a lot of CISOs have been warning about for months. The firm scanned roughly 380,000 publicly accessible web applications built with AI coding tools from Lovable, Replit, Base44, and Netlify. About 5,000 contained sensitive corporate information. Close to 2,000 exposed genuinely private data to anyone who typed in the URL.
What turned up was specific and verified. WIRED and Axios independently confirmed multiple live apps, including internal financial records for a Brazilian bank, a shipping company’s vessel schedules and port assignments, a UK health company’s active clinical trial data, and full unredacted customer service transcripts for a British cabinet supplier. Hospital work assignments with doctors’ PII. A company’s go-to-market strategy deck. Sales records from multiple firms. All sitting on the open web, much of it indexed by Google.
“This is one of the biggest events ever where people are exposing corporate or other sensitive information to anyone in the world,” RedAccess CEO Dor Zvi told WIRED.
How 5,000 apps ended up on the open internet
Zvi’s team found the exposure while conducting shadow AI research for clients. The discovery method was straightforward reconnaissance. No exploit kits, no zero-days. The apps were findable because several vibe coding platforms default to public access. Users have to manually toggle settings to make their apps private. Many never did.
That’s the same failure pattern behind the Amazon S3 bucket crisis of the early 2020s, when misconfigured cloud storage exposed sensitive data from Verizon, the Pentagon, and World Wrestling Entertainment. Default settings plus untrained users equals corporate data on the open internet. The tools changed. The failure mode didn’t.
VentureBeat called it “the new S3 bucket crisis.” That framing is generous. S3 buckets were at least configured by people who knew they were managing infrastructure. Vibe-coded apps are built by employees who, in many cases, don’t realize they’re deploying software at all.
The warning shots were already fired
RedAccess isn’t the first team to flag this. In October 2025, API security firm Escape.tech scanned 5,600 publicly available vibe-coded applications and found more than 2,000 high-impact vulnerabilities, over 400 exposed secrets including API keys and access tokens, and 175 instances of PII exposure containing medical records and bank account numbers. Every single vulnerability sat in a live production system.
In July 2025, Wiz Research discovered a platform-wide authentication bypass in Base44 that allowed anyone to create a verified account on private apps using nothing more than a publicly visible app ID. No password. No technical sophistication required. Wix patched it within 24 hours, but until then, the flaw had affected every app on the platform.
Gartner’s “Predicts 2026“ report, published in December 2025, forecast that prompt-to-app approaches adopted by citizen developers will increase software defects by 2,500% by 2028. Not a typo. Twenty-five-fold. Gartner specifically flagged a new class of defect where AI generates code that compiles and runs correctly but has no awareness of system architecture or business rules. The code works. It just isn’t safe.
Why this lands hardest on finance
The RedAccess findings included internal financial records, sales data, and corporate strategy documents. Those are exactly the kind of assets that flow through the tools finance professionals are most likely to build on a Saturday afternoon.
CFO Dive reported last month that controllers and FP&A leads are spending evenings and weekends building internal tools and reporting dashboards with AI coding platforms. The prototypes work. The problem is everything around them. Nobody is running access reviews on these tools. Nobody is logging who touches the data. Nobody is tracking where the app is hosted or whether it’s pulling from a sandbox or the production GL.
For public companies, this creates a compliance problem with a short fuse. SOX requires access controls, audit trails, and documented data handling for any system that touches financial reporting. A vibe-coded app pulling variance data from your ERP and rendering it in a browser window with no authentication fails on all three counts. Your external auditors will flag it the moment they find it, and the RedAccess research just taught every auditor in the country where to look.
The Cyberhaven 2026 AI Adoption & Risk Report found that 39.7% of all data movements into AI tools involve sensitive information, and 32.3% of ChatGPT usage at work still runs through personal accounts that bypass SSO, centralized logging, and enterprise retention policies. Finance teams aren’t exempt from this behavior. They’re just handling data with higher regulatory stakes when they do it.
Two things worth doing this week. First, ask your team directly: has anyone built a tool, dashboard, or automation using an AI coding platform in the last six months? You need an inventory before you can govern anything. Second, check whether any of those tools connect to production financial data. If they do and they lack authentication, access logging, or version control, they need to come inside your ITGC framework or go offline until they can.
Last week, I made the case that a vibe-coded app you don’t understand is just another black box. This week, 5,000 of those black boxes were found sitting open on the internet. The governance gap showed up … at scale … with receipts.
The Pro edition this week builds an operational response: a Vibe Code Exposure Audit checklist for finance leaders, SOX control mapping for citizen-developed tools, and a decision framework for when a prototype needs to be handed off to engineering or taken offline.
Unlock the full audit checklist & SOX control mapping: