Three documented security incidents in twelve months. One common cause. And a setting almost nobody outside software development has ever heard of.
I teach people how to build real, live websites in a weekend using AI. I use Lovable myself. I genuinely think it is one of the best things to happen to small business owners in years. You no longer need to hire a developer or learn to code just to get a working website or app off the ground.
So this is not an anti-AI piece. It is the piece I wish someone had handed me before I started recommending these tools to beginners without a second thought.
Lovable's promise is simple: describe what you want in plain English, and it builds a working app. Frontend, database, user logins, all of it, without you writing a line of code. For a lot of use cases, it actually delivers that.
But somewhere in that pitch, one detail gets lost. The moment your project needs to store data — customer names, emails, registrations, anything at all — you are no longer just building a website. You are running a database. And a database that is not configured correctly does not just look broken. It quietly leaks.
This is not a hypothetical. It is a documented pattern.
Spring 2025. A vulnerability tracked as CVE-2025-48757 was found affecting Lovable-built apps. Researchers scanned over 1,600 live Lovable projects and found around 170 of them exposing database credentials and API endpoints to anyone who knew where to look. It was reported to Lovable privately in March. A public fix took months.
February 2026. A researcher testing a Lovable-built app used by universities to manage exam questions and student grades found 16 separate security flaws. Six were critical. The app allowed unauthenticated bulk email sending, unauthenticated account deletion, and grade manipulation — all without logging in. It had already been featured on Lovable's own showcase page, with over 100,000 views, before anyone flagged the problem.
April 2026. The most serious one yet. A security researcher discovered that any free Lovable account could access private data belonging to other users' projects: source code, database credentials, AI chat histories, and customer records, across every project created before November 2025. It had reportedly been reported to Lovable about seven weeks earlier and marked as a duplicate rather than escalated.
Three separate incidents, three different causes, one common thread. Broken access control. Independent research from Veracode has found a large share of AI-generated code contains exploitable security flaws. This is a pattern across the whole AI-builder category. Not a one-off bug in one product.
It comes down to one setting almost nobody outside of software development has ever heard of. Row Level Security, or RLS.
Lovable's database layer runs on Supabase, which is itself a solid, widely used piece of infrastructure. But a brand new database table does not automatically know who is allowed to see which rows of data. RLS is the setting that enforces that. A new table ships with RLS off unless someone deliberately turns it on and writes the rules correctly.
The AI is very good at generating something that looks finished and works when you click around it. It is not reliably good at getting this one invisible setting right, and the person using the tool usually has no idea it exists, let alone how to check it.
When Lovable's own security lead responded to one of these incidents, the explanation was that it is "at the discretion of the user" to apply the platform's security recommendations. That is a fair, honest statement. It is also exactly the problem.
This is not a reason to avoid AI builders. It is a reason to know exactly where the risk actually lives.
If you are building a simple website — a business page, a portfolio, something with no logins and no stored customer data — none of this applies to you. There is no database to misconfigure because there is not one.
The moment you add a database — user accounts, a signup form that stores entries, anything with a login — you have entered the territory where these incidents happened. At that point, a few things are worth doing regardless of what tool you are using:
AI builders like Lovable are a genuine step forward for anyone who is not a developer. But "no code required" was never the same promise as "no expertise required." The gap between those two ideas is exactly where these incidents happened.
Cyber Press, "Lovable AI App Builder Reportedly Exposes Thousands of Project Data via API Flaw," April 2026
The Register, "AI-built app on Lovable exposed 18K users, researcher claims," February 2026
Bastion, "Lovable Data Breach April 2026: What Was Exposed & How to Respond," April 2026
Superblocks, "Lovable Vulnerability Explained: How 170+ Apps Were Exposed"
Computing.co.uk, "Lovable flaw exposed source code, credentials and AI chats," April 2026