Sentrint Blog

Working is not the same as safe

Lovable, Bolt, v0 and Cursor all ship apps with the same handful of holes, and none of them throw an error. One guide per tool, per question.

AI-built apps · Checklist

Six security checks you can run on your own SaaS in 15 minutes

The same handful of mistakes shows up in almost every app built fast. Six checks, no tools required, each with the exact command to run and what a failure looks like.

· 8 min read · Gourab Dasgupta

Six checks, none of which need a tool: row-level security off, a policy that allows everything, secrets in NEXT_PUBLIC_ or VITE_ variables, auth that only runs in the browser, .env in git history, and a webhook with no signature check. Numbers one and four cause most of the breaches you read about.
The running order 22 guides, newest first
01 Why AI-generated code keeps shipping XSS, and the four places it hides AI-built apps · 5 min 02 A v0 security checklist for the Next.js app you just exported v0 · 5 min 03 Supabase RLS is not enough on its own. Here is what else you need AI-built apps · 6 min 04 Six security checks you can run on your own SaaS in 15 minutes AI-built apps · 8 min 05 Write a security rules file once, and every prompt inherits it AI-built apps · 5 min 06 Deleting the key from the file does nothing. It is still in your git history AI-built apps · 5 min 07 Your app calls an LLM. Everything your users type is now instructions AI-built apps · 5 min 08 Your paywall is a CSS rule. Here is what it costs and how to fix it AI-built apps · 5 min 09 The package was real, popular and compromised. Your agent installed it anyway AI-built apps · 4 min 10 A Lovable security checklist you can finish in an evening Lovable · 5 min 11 Is v0 safe? The generated code is the whole question v0 · 5 min 12 Is Replit safe? The agent deleted a production database, and that was the smaller problem Replit · 4 min 13 Is Lovable safe? What it secures, and what it leaves to you Lovable · 5 min 14 Is Cursor safe? The editor and the code it writes are two questions Cursor · 5 min 15 Is Bolt safe? The platform is certified. Your app is not covered by that Bolt · 4 min 16 Is Base44 safe? A single flaw exposed every private app on the platform Base44 · 4 min 17 Your AI invented a package name. Someone else registered it AI-built apps · 6 min 18 Every MCP server you added is code you never read Cursor · 5 min 19 Can one of your users read another user's data? Here is how to find out AI-built apps · 6 min 20 A Bolt security checklist for the hour before you share the link Bolt · 5 min 21 Your API key is in your frontend. Here is how to find it and what to do AI-built apps · 5 min 22 What the research actually says about AI-generated code security AI-built apps · 5 min
Lovable 2 guides
Checklist A Lovable security checklist you can finish in an evening Nine checks for a Lovable app before you send it to real users. Each one names what to click, what a failure looks like, and a prompt you can paste straight back into Lovable to fix it. · 5 min read Is it safe? Is Lovable safe? What it secures, and what it leaves to you Lovable handles hosting, HTTPS and infrastructure patching. Authorisation, API keys, database rules and paywalls are yours. Here is the line, and how to check your side of it. · 5 min read
Bolt 2 guides
Is it safe? Is Bolt safe? The platform is certified. Your app is not covered by that Bolt is SOC 2 Type 2 certified, runs your code in an isolated browser sandbox and offers an opt-in security review. None of those cover the code it wrote for you. Here is the line. · 4 min read Checklist A Bolt security checklist for the hour before you share the link Eight checks for a Bolt app, in the order a real attacker would find them. Most take two minutes and need nothing but a browser, a second account and your Supabase dashboard. · 5 min read
v0 2 guides
Checklist A v0 security checklist for the Next.js app you just exported v0 writes idiomatic Next.js, so its failures are Next.js failures. Middleware doing authorisation, server actions with no caller check, and secrets that crossed into a client component. · 5 min read Is it safe? Is v0 safe? The generated code is the whole question v0 produces React and Next.js you own and deploy yourself, which moves more of the security surface onto you than a hosted builder does. What that changes, and the four checks that matter most for a Next.js app. · 5 min read
Cursor 2 guides
Is it safe? Is Cursor safe? The editor and the code it writes are two questions Cursor has had three rounds of remote code execution flaws, all patched, all needing you to open something hostile. The code it writes into your repository is the half nobody audits. · 5 min read Risks Every MCP server you added is code you never read An MCP server describes its own tools, and your agent trusts that description. Two Cursor CVEs came out of exactly that, and the general problem is not patched because it is not a bug. · 5 min read
Replit 1 guide
Is it safe? Is Replit safe? The agent deleted a production database, and that was the smaller problem In July 2025 Replit's agent dropped a live production database during a code freeze and then covered it up. Here is what that incident actually proves, and the five things it does not. · 4 min read
Base44 1 guide
Is it safe? Is Base44 safe? A single flaw exposed every private app on the platform In July 2025 Wiz found that two Base44 endpoints needed no authentication, and an app_id sitting in a public manifest was enough to register into a private, SSO-protected app. Wix patched it in 24 hours. · 4 min read
AI-built apps 12 guides
Vulnerabilities Why AI-generated code keeps shipping XSS, and the four places it hides Georgetown's CSET found cross-site scripting in 86% of AI-generated samples across five major models. React protects you everywhere except the four spots a model reaches for when you ask it to render formatted text. · 5 min read Guide Supabase RLS is not enough on its own. Here is what else you need Turning on Row Level Security and writing a policy leaves the default grants in place. Supabase's own docs say so. What that means for an AI-built app, and the four steps that actually close it. · 6 min read Checklist Six security checks you can run on your own SaaS in 15 minutes The same handful of mistakes shows up in almost every app built fast. Six checks, no tools required, each with the exact command to run and what a failure looks like. · 8 min read Guide Write a security rules file once, and every prompt inherits it CLAUDE.md, AGENTS.md, .cursorrules and .windsurfrules are standing instructions your AI tool reads on every request. A copy-paste security rules file, and the four rules that earn their place in it. · 5 min read How to secure it Deleting the key from the file does nothing. It is still in your git history A secret committed once stays in every clone, every fork and every CI cache until you rotate it. Rotation is the fix. Rewriting history is cleanup, and it comes second. · 5 min read Risks Your app calls an LLM. Everything your users type is now instructions If your app sends user text to a model, the model cannot tell that text apart from your prompt. The fix is not better prompt wording, it is deciding what the model is allowed to reach. · 5 min read Vulnerabilities Your paywall is a CSS rule. Here is what it costs and how to fix it Hiding a premium feature is presentation. Checking a plan is authorisation. Generated apps almost always do the first, and the gap is visible to anyone who opens developer tools. · 5 min read Risks The package was real, popular and compromised. Your agent installed it anyway Shai-Hulud is a worm that steals npm, GitHub, AWS and GCP tokens through postinstall scripts, then republishes itself. In August 2026 it took keyv and reached over 400 packages in hours. · 4 min read Vulnerabilities Your AI invented a package name. Someone else registered it Slopsquatting is the attack where an LLM hallucinates a dependency that does not exist, and an attacker registers that exact name. One fake package reached 237 repositories on its own. How to check yours. · 6 min read Risks Can one of your users read another user's data? Here is how to find out Broken object level authorisation is the most common serious flaw in AI-built apps, and it is invisible when you only ever test your own account. A twenty-minute test, and the fix for each place it hides. · 6 min read Vulnerabilities Your API key is in your frontend. Here is how to find it and what to do If a key reached your browser bundle, it is public, and rotating it is only step two. How to find keys in a deployed app, which ones are meant to be public, and how to move the rest server-side. · 5 min read Guide What the research actually says about AI-generated code security Eleven numbers people quote about AI code security, what each study actually measured, and which of them survive a look at the methodology. Updated as new research lands. · 5 min read