
The app was meant for your team. Anyone with the link could get in
A time-tracking app built with Grok sent passwords for 18 accounts to the browser whenever someone opened the page. They were stored in plain text in a JavaScript file. Some were administrator passwords. Claroc documented the case on 10 November in an investigation that found internal tools people could open without signing in. Claroc’s investigation
On screen, you only see what the app chooses to display. The browser may have received much more. A test password left in an HTML comment, for example, is still part of the page sent to every visitor. Anyone who receives that code can read it. That’s one of the things to check before going live, alongside the files the page loads and the data the server returns. OWASP testing guide
We’ve already looked at how vibe coding can help get a first version off the ground. The step I’m interested in here is when that app starts handling customer data, staff accounts or payments. At that point, you need to know who can see which data and what happens when the business needs to change something. Vibe coding and web development
Signing in with Google is the first step
As part of the same investigation, Claroc used Grok to build a test app for a sales team. It had a ‘Sign in with Google’ button. Someone using an account that hadn’t been invited was able to sign in, see the records and edit them. On sign-in, the code added the user to the team before checking whether they should have access. Claroc’s investigation
Google confirms which account the person used to sign in. From there, the app has to check whether that account belongs to the team, what role it has and what information it can view. Those are business decisions. You might want any customer to be able to sign up while leaving order management, refunds and access to customer records to your staff. Within the team, taking payments, helping customers and administering the app may each require different permissions.
When you ask AI to add a sign-in feature, it’s easy to call the job done once the button works. The job also involves setting access rules and checking that the server applies them every time someone requests information. A valid account should only give you access to the data you’re entitled to see. That rule has to hold even when someone follows a direct link. OWASP on checking access to individual records
Who can open that record?
Imagine a fictional app for camp registrations. A record can list attendance dates, allergies, authorised adults for pick-up and payments already made. If the link reaches another family that also has an account, the app needs to check which registration that account is linked to before showing any data. The same rule should protect any receipt that can be downloaded from the record.
Within the team, the staff handling payments may need to know the amount and whether it’s been paid. A camp leader needs their group’s list and any alerts relevant to looking after it. The camp coordinator may need a broader view to deal with incidents. When the app is built, those differences determine the permissions and the fields the server sends to each account. Hiding a menu item achieves little if those fields still reach the browser.
There are real cases of apps that made this kind of data accessible. In November, Crissy Field documented projects built with Lovable and connected to Supabase where it was possible to read records without signing in. Among the examples were ticketing and food sales systems that exposed email addresses, phone numbers, postal addresses and ticket codes. The report mentions payment information and leaves open whether it included full card numbers or security codes. Crissy Field’s investigation
Lovable describes checks that review code and data access rules before publication. Those warnings are part of the review. Making sense of them takes knowing what the app is meant to show. In the camp example, available places can be public. An individual registration needs different permissions. Lovable’s pre-publication checks
What data does the app handle when someone pays by card?
In May, Sansec observed attacks that exploited a vulnerability in a WooCommerce plugin to inject code capable of capturing card numbers and security codes from the payment form. The vulnerability was in the plugin. The case helps explain why a payment review also looks at where card details are entered and what code runs on that page. Sansec’s investigation
In a redirect integration like the one Redsys documents, the buyer enters their card details on the payment provider’s page. The business’s app still manages the registration, the transaction reference, the amount and the payment status. The server confirms the payment once it receives and validates the provider’s notification. The ‘Thank you’ page is there to inform the buyer. How the Redsys redirect works
Part of the review is checking that the payment is still recorded if the buyer closes the tab, and that two notifications for the same payment produce a single confirmation. Knowing how the payment and the registration are linked also helps if you ever need to restore the app from a backup. How to restore your website without losing orders
What a developer brings before launch
Before adding real data, you need to agree on who can see or change each thing. Once the rules are clear, you can check what data reaches the browser, where credentials are stored and how the app responds when someone using another account requests a record or file. A test environment also lets you check permission changes and payments without affecting customers. You can automate the checks that are worth running again after each change. Why use staging before publishing
I work with AI because it lets me take on projects that were once difficult to fit in because of the time or complexity involved. It can help me compare approaches, generate code and prepare tests. My job is to give it the business context, decide which solution fits and check how it behaves before putting it in users’ hands.
That decision also affects maintenance. When a permission applies in the dashboard, downloads and reports, it helps if all three use the same rule. That gives you one clear place to update it and tests that show what still works. You can also choose technology your team already knows and can maintain, provided it fits what the project needs.
Who can take over the project after handover?
Lovable allows you to export your code and host it outside the platform. The database and uploaded files are transferred separately. To know whether someone else can take over, you also need to know who controls the domain and service accounts, how to recover the data and what it takes to deploy a change. Lovable’s documentation on hosting and ownership
If someone offers you a website for €400, the price alone tells you very little about what you’ll receive. A page that lists opening hours and an app that manages registrations, permissions and payments are different jobs. Ask what the handover includes and who can carry on the work when a price changes, someone new joins the team or you need to restore a backup.
If you already have an app built with AI and want to open it to your team or customers, we can review how it handles access, payments and maintenance. Tell me what you’re building
Share this article