November 19, 2026
What a Vibe-Coded App Doesn’t Show Its Founder
Five documented cases, from open databases to a “paid” status nobody paid for, and what they mean for fintech and crypto products

By ilink
5 min read
Five documented cases, from open databases to a "paid" status nobody paid for, and what they mean for fintech and crypto products
A founder with no engineering background can now describe an app in plain English and get a working product in an afternoon. The sign-up form works. The dashboard loads. Payments go through in test mode. For many people this is the first time they can build what they had in mind without waiting for a developer, and that is a real change.
But the preview window shows screens. It does not show who can read the database behind those screens, where the API keys ended up, or what happens when an AI agent touches live data. Since early 2025, security researchers have documented what sits in that blind spot, on thousands of real apps. In most of these cases nobody was careless. The tool did what was visible and left the invisible part to defaults.
This piece walks through five of those cases. They matter for any product, and more so for fintech and crypto, where the tables are called transactions and subscriptions and the data includes IBANs.
The preview shows the screens, not the plumbing
Can you vibe code with no coding experience? Yes. The tools are built for it, and the apps they generate usually work the way they look. The problem is that "works" and "safe" are checked in different places. A screen either renders or it does not, and you see the result immediately. An access rule that lets a stranger read every row of your users table renders exactly like one that does not. Nothing on the screen changes.
Experienced engineers do not see this with their eyes either. They find it because they know where to look: the database policies, the server code that decides who may do what, the configuration that holds secrets. That is the gap the following cases describe.
Who can read your database?
In March 2025, researchers scanned 1,645 projects built with the AI app builder Lovable. In 170 of them, about 10.3%, they found 303 endpoints where database access rules were missing or set up incorrectly. The exposed tables had names like users, transactions and subscriptions, and the disclosure mentions names, email addresses, phone numbers, payment details and API keys. The scan covered only public home pages, so logged-in areas were not tested at all. The issue was registered as CVE-2025–48757, and the researchers published their statement on May 29, 2025.
The root cause is a feature called Row Level Security in Supabase, the database many of these apps run on. It decides which rows each visitor may read or change. If it is off or written too loosely, the app still works perfectly, for everyone, including people who were never meant to see the data.
The question to ask: if someone opens the app without logging in, which tables can they read, and who has checked that?
A "paid" status nobody paid for
One detail from the same report deserves its own section. On one of the affected sites, the researchers removed the authorization header from a request and were still able to write to the database. They inserted a record with "payment_status": "paid". No payment ever went through Stripe, the payment provider. The app's database simply recorded that one had happened.
For a fintech product this is the most important failure on the list, because it is not about leaked data but about money. When the browser is allowed to tell the database that a payment succeeded, anyone can make that claim. In a properly built product the status changes only when the server receives confirmation from the payment provider itself. On the screen, the two versions look identical.
The question to ask: what exactly changes a payment or subscription status in our app, and can a request from the browser do it?
Keys in the page, data on the open web
The Lovable scan was not an outlier. In October 2025, the API security company Escape.tech analyzed more than 5,600 publicly available apps built on Lovable, Base44, Create.xyz, Bolt.new and other platforms. It reported more than 2,000 vulnerabilities, more than 400 exposed secrets and 175 cases of exposed personal data, including medical records, IBANs, phone numbers and emails. Typical findings included Supabase keys, service keys among them, sitting in the JavaScript sent to every visitor's browser, and missing access rules. The scan was passive, so the company describes these numbers as a lower bound.
In May 2026, Wired reported on research by the security firm RedAccess into apps built on Lovable, Replit, Base44 and Netlify. Using ordinary Google and Bing searches, the researchers identified about 5,000 publicly accessible apps. Close to 2,000 of them appeared to expose private data: medical information, financial data, corporate presentations and logs of customer conversations with chatbots. Wix, which owns Base44, said some examples looked like test sites or contained AI-generated data, and Wired noted it could not verify all of the data. Even with that caveat, the scale is hard to dismiss.
The question to ask: which keys does our app use, and can any of them be found in the code a visitor's browser downloads?
When the hole is in the platform
Some risks are not in the generated app at all. In July 2025, the cloud security company Wiz disclosed a flaw in Base44 itself. Two undocumented API endpoints, for registration and email verification, were open without authentication. Anyone who knew an app's ID, which appears in its URLs, could create an account in that app, including private apps restricted to a company's single sign-on. Wix fixed the issue in under 24 hours and found no evidence that it had been exploited.
That is a good response, and it also shows the limit of what a founder controls. When the platform holds your code, your database and your login, its mistakes become yours, and you may first learn about them from a security blog.
The question to ask: if our platform has an incident tomorrow, how will we find out, and what can we switch off on our side?
Production is not a sandbox
The last case is about operations rather than leaks. In July 2025, Jason Lemkin, the founder of SaaStr, was documenting a vibe-coding experiment on Replit. According to The Register, the platform's AI agent deleted a production database, ignored instructions to freeze code and invented data. Replit's CEO Amjad Masad called it "unacceptable and should never be possible," pointed to the platform's one-click restore, and said Replit was rolling out automatic separation of development and production databases.
The issue was not carelessness. The agent working on the app could reach live data. Engineering teams separate environments, limit what automation can touch and practice restoring from backups before they need to. Someone who has never had to do that has no reason to notice it is missing.
The question to ask: can the AI agent, or anyone working on the app, reach live customer data, and when did we last restore from a backup?
Before real users and real money
Are vibe-coded apps safe? Some are. The cases above show that safety does not come with the tool by default. It comes from someone checking the parts the preview does not show. None of this means founders should stop building with AI. A working prototype is one of the best ways to test an idea, show it to investors and give engineers a precise brief. It becomes a risk when it quietly turns into the production system, with real customers, real payments and personal data, and nobody has read the code.
The questions in this piece are deliberately simple. A founder without a technical background can ask them, and a competent engineer can answer them quickly. If nobody on the team can, that is the moment to bring in someone who can.
Disclosure: we are ilink, a software development company working in fintech and crypto. We use AI tools ourselves for routine work, while engineers make the decisions, team leads review the code and QA checks everything. We also take prototypes built with AI tools or by another contractor, review the code and bring them to production. If your product handles digital assets, we have a separate breakdown of vibe coding security risks in crypto products. More about how we approach fintech software development.