App security for people who build with AI.
This section is for people who built an app with Newly or another AI builder and want to know what to check. Each article answers one question in plain English, lists dated fixed versions where they exist, and marks every finding as detected, unknown or confirmed by owner evidence, so you can see what is proven and what is still open.
Last updated
What to do this week
- 01
Find the keys your app ships with.
List every key in your app's config. Move secret keys to a server and rotate any that were ever in the app.
- 02
Check who can read your database.
Turn on row level security for every Supabase table, or review the equivalent rules in your database, and read each policy.
- 03
Update Node.js wherever it runs.
The fixed releases from October 21, 2026 are 22.23.4, 24.22.1 and 26.12.1. Rebuild and redeploy after updating.
- 04
Ask your host about PHP.
If your website runs on PHP, ask whether the PHP serving it is 8.5.12, 8.4.27, 8.3.36 or 8.2.35, or carries the fix backported.
- 05
Run the 15-minute check.
Mark each item detected, unknown or confirmed by owner evidence, and keep the evidence.
Apps built with AI
What to check in an app an AI builder wrote for you: keys, database rules, auth and AI features.
Is my AI-built app secure? A 15-minute check
Built an app with Newly or another AI builder? Check six things in 15 minutes: bundle keys, Supabase RLS, route auth, AI chat, runtimes and your website.
Updated 7 min read
Pre-launch security review for vibe-coded apps
A pre-launch checklist for vibe-coded apps: keys in the bundle, database rules, route checks, AI feature limits, store rules and the proof you can show.
Updated 9 min read
When the security scanner is the attack: the Trivy action compromise
In March 2026 an attacker moved 76 of 77 trivy-action tags onto credential-stealing code. What happened, why a tag was enough, and how to check your own CI.
Updated 9 min read
Slopsquatting: AI-Hallucinated Packages That Install Malware
Slopsquatting is when attackers register package names that AI assistants invent. See how it reaches an Expo app and what to check in your own project.
Updated 8 min read
Backdoored Models in Coding Agents: Provenance and Egress
A fine-tuned 7B open model made a coding agent send out dummy credentials in a lab demo. What to check on model source, sandbox and outbound access in your app.
Updated 9 min read
Hardening AI Coding Agents in GitHub Actions: A Checklist
Checklist for running Claude Code and other AI agents in GitHub Actions: limit triggers, shrink GITHUB_TOKEN, restrict egress and keep secrets out of the job.
Updated 8 min read
PixelLeak: AI Agents Posted 13,000 Screenshots on GitHub
Glow Labs found AI coding agents hosting 13,000+ internal screenshots in public GitHub repos. What leaked, why, and how to check your own agent setup and repos.
Updated 8 min read
Backend, auth and keys
Database rules, sign-in middleware, environment variables and API keys: what an AI-built backend usually gets wrong and how to check it.
Vercel Security Incident: Which Env Vars to Rotate
Vercel closed its April 2026 incident on October 16. Here is which unmarked env vars an app backend should rotate, in what order, and how to check for misuse.
Updated 8 min read
Leaked OpenAI API Key on GitHub: How to Check and Rotate
Found an OpenAI or other AI API key in a repo or app bundle? Where AI keys leak, how to check your own project, and how to revoke and replace one that is out.
Updated 9 min read
How to rotate app secrets after a platform breach (Railway, Zeabur)
Backend on Railway or Zeabur? Replace every stored key at its issuer, redeploy, revoke the old one, then read the logs. Vendor facts checked 24 October 2026.
Updated 7 min read
Supabase anon key exposed: what it allows and how to test RLS
Your Supabase anon or publishable key is public by design. Whether it opens your tables depends on RLS. Test every table as anon and see what each result means.
Updated 8 min read
Clerk createRouteMatcher bypass: middleware-only auth isn't enough
CVE-2026-41248 let crafted requests skip Clerk's route matcher. See which versions are affected, how to fix them, and how to check your own routes for middleware-only auth.
Updated 9 min read
Firebase Security Rules Audit: the Firewreck Pattern in 2026
Find open Firebase rules before a stranger does: three patterns behind the 2026 Firewreck findings, how to test them in your app, and what each result means.
Updated 9 min read
Better Auth June 2026 Security Update: OAuth, SSO and SCIM
Better Auth fixed 13 advisories in June 2026 across OAuth, SSO and SCIM, with newer ones since. Which versions to run and what to check in your own app.
Updated 9 min read
Scan your live site
red.run.cloud's free scan matches your site's stack to known vulnerabilities, one free scan per domain. The $100 full scan sends an AI red team that looks for a path an outsider could use, proves access once, stops, and reports every step.
You sign in with Google, GitHub or an emailed code, and verify your domain with one DNS TXT record. The red team never deletes, changes or downloads your data.
A scan reports what it detected from outside. A matched version number cannot show a fix your host backported, so treat a clean result as unknown until you confirm it.