Sites built on AI platforms expose confidential data, RedAccess study finds

AI website-building platforms such as Lovable, Base44, Replit and Netlify promise that anyone can become a developer in seconds. But an analysis by RedAccess, a cybersecurity company, uncovered a large-scale problem: thousands of deployed sites have passwords, API keys and other confidential data sitting in plain view.

Scale of the leaks: thousands of sites compromised

Lovable alone hosts more than 20 thousand public sites. When RedAccess researchers analyzed a significant share of the projects, they found a systematic and serious problem: users often put confidential information into prompts, and it then ends up in the generated code. They often do not realize that data left in examples, comments or even pasted accidentally from the clipboard will be visible to anyone who opens the site's source code.

The problem is made worse by the fact that the platforms deliberately publish sites at publicly accessible URLs, and the source code is often available directly in the browser's view-source tool.

RedAccess found the following in the open:

• Passwords and API tokens in source code
• Keys for cloud services (AWS, Google Cloud, Azure)
• Email addresses and mobile numbers of employees
• Links to internal admin panels and services
• Database dumps containing customers' personal information
• SSH logins and passwords for remote server access

Platforms put the blame on users

The developers of Lovable, Base44, Replit and Netlify take a common position: they provide the tools, and users are responsible for the content they put into those tools. Formally, that is fair, just as a knife maker is not responsible if someone gets cut with a knife.

But security experts point to a critical difference: these platforms are explicitly designed and marketed for people with no coding experience. When a system targets beginners who, by definition, do not know what an API key is or why it must not be left in code, the developer should either add a very prominent warning or build in automatic scanning and blocking of sensitive data before publication.

"The platforms are designed for people who do not know the basics of information security.

When a system is built for them, its creator bears responsibility for basic protection," the RedAccess study says.

What this means for the industry

AI generators are literally changing the landscape of web development: they are several orders of magnitude cheaper and faster than hiring an experienced developer. But they create a new category of security problems that the older generation of developers simply did not anticipate.

For companies using such services, the advice is simple: if you use Lovable, Replit or similar platforms, review the generated code before publishing to production. Never give AI systems real passwords, API keys or access to confidential data. Use fixed placeholder values.

For the platforms themselves, this is a window of opportunity. The first company to add built-in automatic code scanning for data leaks and to block deployment when sensitive information is detected will gain a huge competitive advantage in the form of trust from corporate and government customers. Otherwise, RedAccess will remain just a story about a time when the industry was still careless.