Securing apps built with AI coding tools
Practical, sourced answers to the security questions founders hit after shipping with AI coding tools and app builders: who can read whose data, where the keys went, what an AI feature can be talked into, and how to show a buyer what you checked.
- Practice8 min read
Is vibe coding secure? What the research says, and what to check
Vibe coding does not make an app insecure by itself, but it does not make it secure either. What the studies of AI generated code found, where the flaws cluster, and what to check before real users arrive.
Read - Guide9 min read
How to secure an app made with an app builder: start with the database rules
Apps from browser builders often read a hosted database straight from the browser with a key that is public by design. The database rules are the only gate. How row level security works, how to write the policies, and how to test them as a stranger.
Read - Guide8 min read
Your API key is in the frontend. How to tell, and what to do next
Anything shipped to the browser is public, including keys in your JavaScript bundle or in public environment variables. How to check your own app in ten minutes, the order to fix it in, and how to call a paid API without handing out the key.
Read - Guide8 min read
IDOR in AI-built apps: how one changed number exposes another user’s data
An insecure direct object reference is the most common shape of broken access control, which OWASP ranks first. What it looks like, why AI coding tools write it, why random ids do not fix it, and how to find and close it.
Read - AI security8 min read
How to prevent prompt injection in your app’s AI feature
You cannot filter prompt injection out completely, because a language model does not reliably separate instructions from data. You can limit what a successful injection achieves. What the guidance says, and what to change in a typical AI feature.
Read - AI security9 min read
How to secure an MCP server: the attacks the specification warns about
An MCP server lets an AI agent act for a user, so it inherits every API risk plus a few of its own. The attacks the protocol’s own security guidance describes, and the design choices that close them.
Read - Practice9 min read
How to answer a security questionnaire when you are a small startup
The first enterprise deal usually arrives with a spreadsheet of security questions. You do not need an audit report to answer it well. What buyers are really checking, how to write each answer, what counts as evidence, and what never to write.
Read - Method7 min read
Why VibeLock will never give you a security score
Something you show an investor has to survive being questioned. That rules out asking a model to grade you, and followed far enough it rules out any number of our own invention.
Read - Practice9 min read
What AI coding tools actually get wrong, and what they do not
The security failures in AI generated applications are not exotic. They are a small, predictable set, and knowing which ones lets you check the right twenty things instead of the wrong two hundred.
Read - Compliance8 min read
The EU AI Act when you are three people and an app
Most of the writing about the AI Act is addressed to organisations with a compliance function. Here is the part that applies when you do not have one, and the part that probably does not apply to you at all.
Read - Practice6 min read
The security question in the fundraise, and how it actually goes
Nobody is expecting a seed stage company to have SOC 2. What they are testing is whether you can tell the difference between what you have done and what you have not.
Read