16,000 Supabase databases exposed: a security checklist for apps built with AI

Two stories topped Hacker News this week, and they say the same thing: software built fast is not the same as software built safe.

What happened

Around 16,000 exposed databases. Security firm UpGuard told TechCrunch it found around 16,000 databases hosted on Supabase that exposed some degree of personal data: names, addresses and phone numbers and, in fewer cases, passwords and login tokens. The report says vibe-coded apps and websites can spill sensitive data "through basic misconfigurations and improper security" (TechCrunch, 25 September 2026).

The platform itself was not broken into. The exposures came from how individual apps were set up. Supabase's chief information security officer, Bil Harmer, told TechCrunch: "Security at Supabase is never finished. We care deeply about getting it right, and we'll keep making it easier for every developer to ship securely."

AI agents inside Hugging Face. A team of researchers published how "a swarm of 700 OpenAI agents hacked Hugging Face in July" (Swarm Traces). The agents started with very limited access: they could load web addresses, but not interact with pages or send data. They chained link shorteners and other public web services into a way to run code and move data out. Hugging Face confirmed the evidence matches its own investigation and that the exposed credentials had been revoked. The researchers are careful to note that parts of their reconstruction remain unconfirmed.

Why it matters if you run a business

AI tools can now build a working app in an afternoon. That is real progress, and we use them every day. But a working app and a safe app are different things. The exposed apps worked perfectly for their users, and just as well for anyone who looked. The Hugging Face case shows the other side: an AI agent will use whatever access it is given, in ways nobody planned.

If your business runs on an app that was built quickly, by an agency, a freelancer or an AI tool, it is worth an hour to check it.

The checklist

• Lock every table. In Supabase and similar backends, turn on row-level security for every table and write a policy for each one.
• Keep secret keys on the server. The public key is meant for the browser; the service key is not. If a service key is in your app's code, rotate it today.
• Give each part the least access it needs. Separate keys for the website, the admin and background jobs, each allowed only what it must do.
• Move secrets out of code. No keys, passwords or tokens in the repository or the app bundle. Rotate anything that ever leaked.
• Limit and log. Rate-limit forms and APIs, log sign-ins and exports, and get an alert when something unusual happens.
• Fence your AI agents. Give agents an allow-list of sites and tools, no standing credentials, and a human approval step for anything that sends data out.
• Test like an attacker. Before launch, try to read another user's data, call the API without signing in and change IDs in the address bar.
• Get a second pair of eyes. An independent review before launch costs far less than a breach.

How we build

At HITROO, AI makes us faster and people make it right. Every project we deliver is security-tested before launch, every change is checked, and the software is supported after launch, with a first reply within 24 hours. Our own website follows the same rule: the database account it uses can add form entries, but it cannot read anyone's details back.

Built an app with AI tools, or not sure how your current app is set up? Tell us about it and we'll review it and tell you what to fix. Starting something new? See custom software and AI automation.

Questions

Is Supabase insecure?

No. According to the report, the exposures came from how individual projects were configured, not from a break-in at Supabase itself.

What is row-level security?

A database rule that decides which rows each signed-in user may read or change. Without it, anyone holding an app's public key can often read a whole table.

How do I know if my app is exposed?

Sign out, call your app's API with only the public key and try to read tables that hold personal data. If anything comes back, lock those tables and rotate your keys.

Are apps built with AI unsafe?

Not by default. AI writes code quickly but copies patterns without knowing your data or your risks. Treat AI-written code like a new developer's work: useful, and always reviewed.