AI app builders and coding assistants are very good at producing an app that works. They're less reliable at producing one that's safe. In Veracode's 2025 study of more than 100 language models across 80 coding tasks, AI-generated code introduced a security flaw in 45% of cases. The app runs, the demo looks great, and the hole only shows up when someone goes looking for it.
These are the five problems we look for in every app we move, why AI-built apps are prone to them, and how to check your own app.
1. Keys and secrets exposed in the front end
To make a feature work, an AI assistant will often put an API key wherever the code that needs it lives, and in a single-page app that's usually the browser. Anything shipped to the browser can be read by anyone who opens the developer tools. An exposed AI or email key can run up your bill. An exposed database admin key can hand over all of your data.
How to check: search your code for keys and look at what your live site sends to the browser. Our guide to API keys in your front end walks through it.
2. Data that anyone can read
Many AI builders put your data in a database that the browser talks to directly, such as Supabase. That's safe only if the database's access rules stop each visitor from reading or changing other people's records. When those rules are missing or too loose, anyone with your app's public key can query the tables. This isn't hypothetical: in 2025 a security researcher reported Lovable-built apps whose database rules let unauthenticated visitors read or write their tables, recorded as CVE-2025-48757. Lovable disputes the record, saying each customer is responsible for protecting their app's data.
How to check: confirm every table has access rules and test them while signed out. See our guide to Supabase Row Level Security.
3. Sign-in and access rules that only live in the browser
Hiding the admin button from ordinary users isn't the same as stopping them using the admin features. AI-generated apps often check roles in the interface and nowhere else, so the server or database will happily do whatever the browser asks. Every rule about who can see or change something has to be enforced on the server or in the database.
How to check: sign in as an ordinary user and try the actions only admins should have. Change a record ID in a request and see whether you get someone else's data.
4. Outdated, or invented, packages
AI tools pull in packages freely and rarely come back to update them, so known vulnerabilities pile up. There's a newer risk too. A study accepted at USENIX Security 2025 found that code-generating models recommend packages that don't exist: at least 5.2% of the time for commercial models and 21.7% for open-source ones, across 205,474 unique made-up names. Attackers can register those names with malicious code and wait for someone to install them.
How to check: run npm audit or your language's equivalent, and look up any package you don't recognise before you trust it.
5. AI agents with write access to production
An AI agent that can edit code can often edit data too, and the line between your test data and your live data isn't always obvious. Replit has described a 2025 incident where an AI agent deleted a customer's database data, which was restored by rollback. Keep agents on development data and keep your own backups. Our guide to AI coding agents and production data covers the guardrails.
In July 2025 Wiz Research reported a flaw in Base44 itself that let outsiders create accounts on private apps using only the app's ID. Base44 fixed it within days and reported no evidence of misuse. On a builder, you rely on the platform's security as well as your own, and you can't patch or audit it yourself.
What AppExit checks on every migration
Every migration is reviewed and carried out by developers, not left to an AI. Before and after the move, we look for exposed keys, data that anyone can read, sign-in and access rules, and outdated packages with known problems, and we tell you what we find in plain English. Anything the move itself breaks, we fix before handoff.
Moving also closes some gaps by design. We take an independent backup before any move, use least-privilege access and review what can be revoked at handoff, and put the hosting, database and domain accounts in your business's name unless you choose our managed hosting.
Web apps from Base44, Lovable, Bolt.new, Replit and Emergent are a flat $199, code and database included. Start your order, or read how we handle access.
Sources
Sources checked November 2026. AppExit is not affiliated with any platform or company named. Brand names belong to their owners.