Researcher scans 159,041 Lovable apps, finds passport numbers and 200K chat messages open to anyone
- Of 159,041 Lovable projects crawled from Common Crawl between December 2025 and September 2026, 34,545 used Supabase, and 24,664 of those sent database records on page load with no login.
- Extra read-only requests with the same API key returned email addresses in 10,560 tables across 7,403 of the 24,664 sites (30%), a figure the author says is not a confirmed leak rate because emails can be public contact or test data.
- A German lifestyle forum exposed more than 80,000 profiles with full names, email addresses, dates of birth and gender, plus more than 200,000 chat messages.
- One service provider exposed more than 1,000 records holding passport numbers, expiration dates and nationalities, and other sites leaked contracts, invoices, photo albums and internal WhatsApp messages.
- Lovable's marketing says no technical knowledge is needed while its terms put data responsibility on customers, and the author says no advanced techniques were used: the team never got past basic requests because so much was already open.
Hacker News 의견들
Nothing to see here, working as intended. Imagine frontier abliterated models running at 2,400 TPS finding all of these in minutes. You are not ready security-wise when that happens.
Or it's just 20 years of move fast and break things catching up, a house of cards that was going to fall in on itself anyway. Formal verification doesn't fix everything, but treating PII as toxic waste instead of treasure to collect leaves you in a much better place.
None of it saves you if the human is completely out of the loop.
I can't state my disdain enough for this. Two decades of learning to build safe software, replaced by this, and the users pay for it.
The standards gap with civil or mechanical engineering is wild. The old joke was that the first woodpecker would destroy civilization, in the AI era a butterfly is enough.
I'm interviewing at a small company going all in on agentic development for a VP of engineering role. I brought up governance, guardrails and the lethal trifecta and got crickets. They said they want a big AI booster. Bullet dodged.
Come on, companies never cared about safe software before AI either. They cared about KPIs.
Pro-AI people aren't a monolith. Learning to use LLMs safely is like learning AWS in 2015 even though people left buckets open. Don't confuse negligent misuse of the tool with the tool itself.
You won't lose your job to AI, you'll lose it to someone using AI, and they'll lose everything to someone who knows security.
Railway got hit by basically this a few days ago, something like ~$740k in AI charges off stolen keys. If it happens to them, random Lovable apps are easy targets.
Breaches existed long before AI, sure. The difference is these are trivial targets, there are tons of them, and the average user has no idea what they're using.
I wrote this post. Submitted it two weeks ago and it didn't get much traction, glad it got resubmitted. Happy to answer questions.
How many of the 159k analyzed sites actually leaked private data?
Author here: no advanced hacking was used, anyone with a moderate understanding of how the web works can reproduce it. We planned to move on to more sophisticated attacks but never got past the basics because so much was already open.
Makes me wonder if Retool's governance-focused pitch is actually valid.
People have basically no understanding of data, privacy and security. Letting anyone build and host a public app is like letting someone fly a passenger plane without a license.
Is it though? How many hundreds of people died, how much infrastructure got destroyed? Security already has a maximalist problem and it doesn't make systems of consequence safer.
Careful what you wish for, next thing you need a government-issued license before a teenager can play with a computer.
Nobody says that about plumbing, heating or electricity.