Security review for AI-built apps
Built it fast with AI tools? Make it safe before launch, a raise or a sale.
An app built with Lovable, Bolt, Cursor, Replit, v0 or a similar tool usually works on the happy path and fails on the paths nobody tried: who can read whose data, where the keys live, what happens at a thousand requests a minute. Our security review finds those gaps in 1 to 3 weeks. Every finding is reproduced, ranked and fixed, every fix gets a test, and a second pass confirms it, before a launch, a raise or a sale.
Before a launch, a raise or a sale.
- Length
- 1 to 3 weeks
- Price
- Quoted to scope
- You get
- Findings, fixes and a retest
Built with
For apps built with AI coding tools.
- Lovable
- Bolt
- Cursor
- Replit
- v0
- Claude Code
- Windsurf
- GitHub Copilot
- and similar tools
These tools are very good at producing something that works when everything goes right. Security is about what happens when something goes wrong, or when someone tries to make it go wrong, and that is the part a prompt rarely asks for.
What we find
What AI tools usually miss.
Data access rules
Row-level rules missing or too broad, so one signed-in user can read another’s records through the API.
Keys in the wrong place
Secret keys shipped to the browser, committed to the repository, or shared across environments.
Sign-in and recovery
Account recovery, invitation and email change flows that can be abused to take over an account.
No limits
No rate limits or lockouts, so a form, a login or an AI feature can be hammered or run up a large bill.
What leaves for AI providers
Customer data sent to model providers without anyone checking what they store, or prompts that can be steered to leak it.
Backups nobody has restored
Backups that exist on paper, with no proven restore, and access that outlives the person who needed it.
Process
How a review runs, step by step.
- 01
Scope
A signed scope, agreed by both sides, before any testing starts.
- 02
Sweep
Agents probe sign-in, access, limits, stored keys, infrastructure and outside services, under supervision.
- 03
Report
Every finding reproduced, ranked by severity and explained in terms your team can act on.
- 04
Fix
We fix what we find within the agreed price, or guide your team, and every fix gets a test.
- 05
Retest
A second pass confirms each fix, and a regression test guards against it returning.
Outcome
What you have at the end.
- A written report your investors, buyers or customers can read
- Fixes for what we find, each with a test
- A retest confirming each fix
- A short list of what to keep watching as you grow
Questions
Questions about the security review.
Do we have to rebuild the app?
Rarely. Most AI-built apps need targeted fixes rather than a rewrite: tighter access rules, keys moved server-side, limits added, flows hardened. If a rebuild is the honest answer, we will say so and price it.
Will you need access to our code and cloud accounts?
Read access to the repository and the hosting and database accounts in scope, granted for the review and removed at the end. Nothing outside the signed scope is tested.
Is this enough for investor due diligence?
It gives you a credible, written answer to the technical security questions in due diligence, with evidence of fixes. If an investor or customer asks for a certified penetration test, we will tell you and work alongside the firm you choose.
Is the security review a penetration test?
No. It is an engineering-led review, with every finding reproduced, fixed and retested. If a customer or regulator needs a certified penetration test, we will say so and work alongside the firm you choose.
Read more
- How to make an AI-built app safe before launch
- How to check Supabase row-level security
- Penetration test or security review: which you need
- When to move off Lovable, Bubble or a similar tool
- Security and hardening for any codebase
- How to keep AI agents away from production
- Turning a prototype into production software
About Rare Agentics
- What we do
- We build bespoke software, custom AI agents and operating platforms, and review and harden software built fast with AI tools.
- Who for
- Founders and growing businesses in any sector, from a London base.
- How we price
- Scoping is free, every piece of paid work has a fixed price in writing before it starts, and a proof sprint starts from £3,000.
- How we control agents
- Senior engineers lead every build and AI agents do the heavy lifting. Agents cannot reach production, and a person signs off every release.
- Where code lives
- Your code lives in your repositories and your data in your cloud accounts from the first commit.
Start a conversation
Send us the app. We will tell you what we would check first.
Send a brief. It does not need to be technical. Within two working days you get a first view, and within a week a written plan, a fixed price and the risks.