A new security investigation has uncovered roughly 16,000 Supabase databases containing potentially sensitive information that was accessible from the public internet — raising fresh concerns about the security risks surrounding the rapid rise of AI-built apps.
The booming popularity of so-called “vibe coding” is making it easier than ever for people with limited software development experience to build websites and applications.
But a new security investigation suggests that convenience can come with a serious cost.
Cybersecurity firm UpGuard says it identified around 16,000 databases hosted by Supabase where some level of personal information was exposed to the public web.
The exposed information included names, addresses, phone numbers and, in a smaller number of cases, passwords and authentication tokens.
Supabase is a popular backend platform that allows developers to store data and operate databases for websites and applications. The company has benefited significantly from the recent surge in AI-assisted software development, particularly as developers use generative AI tools to rapidly create and deploy applications.
Earlier this year, Supabase reached a reported $10 billion valuation.
Its growing role in the AI development ecosystem, however, has also brought greater scrutiny of the security practices surrounding applications built on the platform.
How Supabase Databases Can Become Exposed
The security issue is not necessarily the result of a vulnerability in Supabase itself.
Instead, many incidents involve developers incorrectly configuring their databases or failing to understand which information their applications are making publicly accessible.
That distinction is important.
Platforms such as Supabase provide infrastructure and security tools, but developers remain responsible for configuring their individual projects correctly and controlling who can access the underlying data.
For experienced developers, database permissions and access controls are familiar security considerations.
For people building applications with AI coding tools, however, those requirements can be easier to overlook.
An AI system may generate functional code without necessarily ensuring that every database permission, authentication rule or access policy is configured appropriately for a production environment.
The result can be an application that appears to work perfectly while quietly exposing information that should remain private.
What UpGuard Found in the Exposed Databases
UpGuard said its investigation was designed to assess the scale of publicly accessible data across Supabase-hosted projects.
The company found exposed information including names, addresses, telephone numbers and user passwords. It also identified a smaller number of authentication tokens and other sensitive credentials.
The databases were associated with a wide range of applications and organisations.
According to UpGuard, some of the exposed information included private conversations involving sex workers on an Indian adult streaming platform, thousands of vehicle license plates belonging to users of a U.S. valet service and contact information associated with an immigration and relocation company.
One database was reportedly connected to an African government’s consulate in France.
Another was being used by a virtual SIM operation to intercept text messages containing one-time verification codes.
Such systems can be particularly concerning because one-time passcodes are commonly used as part of account authentication. Intercepting those messages can potentially enable abuse of online account-verification systems.
UpGuard said the majority of the exposed datasets it identified appeared to be located in the United States, although the problem was not limited to one country or region.
AI “Vibe Coding” Is Creating a New Security Challenge
The findings arrive as AI-assisted programming becomes increasingly mainstream.
Tools powered by generative AI can allow users to describe an application in ordinary language and receive working code in return.
This approach, often called vibe coding, has lowered the barrier to creating websites and applications.
But building an application and building a secure application are two different things.
An AI-generated application may contain vulnerabilities in its code, while the person deploying it may not have enough security knowledge to identify the problem.
Configuration is another major concern.
A developer might successfully create a database, connect it to an application and make everything function correctly without realising that a particular table or API endpoint is accessible to anyone on the internet.
That creates a dangerous gap between functionality and security.
The issue is not unique to AI-generated software. Misconfigured databases and cloud storage have been responsible for data exposures for years.
Previous incidents involving improperly secured systems have exposed everything from government files and military communications to immigration records, driver’s license scans and children’s personal information.
What is changing is the scale and speed at which inexperienced developers can now create and deploy software.
Supabase Has Faced Scrutiny as Its User Base Grows
Supabase has introduced changes to its platform over time aimed at strengthening database security and improving how users manage access to their projects.
The company has also expanded security features and tools intended to help developers protect their databases.
When contacted about the UpGuard research, Supabase Chief Information Security Officer Bil Harmer said the company had not seen the research but maintained that its projects are “secure by default.”
Harmer also described security as a shared responsibility between Supabase and its customers.
The company provides security defaults and tools, while customers determine how their individual projects are configured.
According to Harmer, Supabase also notifies customers when security issues affecting their projects are identified.
“Security at Supabase is never finished,” Harmer said, adding that the company intends to make it easier for developers to deploy applications securely.
That distinction between platform security and customer configuration is central to the debate surrounding exposed databases.
Even a platform with strong security controls cannot necessarily prevent a customer from deliberately or accidentally making information publicly accessible.
The Problem Goes Beyond Supabase
The latest findings should not be interpreted as evidence that Supabase is uniquely responsible for exposed databases.
Poorly configured databases, cloud storage systems and web applications have been a recurring cybersecurity problem across the technology industry for years.
What makes the current situation particularly notable is the intersection of that longstanding problem with the rapid growth of AI-assisted development.
Earlier investigations have already identified publicly accessible Supabase databases belonging to startups and applications, including companies associated with Y Combinator.
As more developers use platforms such as Supabase to quickly launch AI-generated applications, the number of databases being created — and potentially misconfigured — can also increase.
That creates a security challenge that involves developers, AI coding tools and infrastructure providers alike.
A Warning for the Next Generation of AI-Built Apps
The UpGuard investigation highlights a broader lesson about the current AI software boom.
Generative AI has made it dramatically easier to create software, but it has not eliminated the need for human oversight.
Database permissions, authentication, access controls and privacy protections still require careful attention, particularly when applications handle personal information.
For developers, a database being functional does not mean it is secure.
For companies deploying AI-generated applications, security testing needs to happen before sensitive information is exposed to real users.
And for infrastructure providers, making secure configurations easier for inexperienced developers could become increasingly important as the number of AI-built applications continues to grow.
UpGuard researcher Greg Pollock said the firm’s research was intended to raise awareness about the scale and significance of publicly exposed data.
The findings serve as another reminder that the biggest security risks in the AI era may not always involve sophisticated hackers or previously unknown vulnerabilities.
Sometimes, the problem can be much simpler: a database that was never meant to be public was accidentally left open to the internet.
As AI makes software development faster and more accessible, ensuring that security keeps pace with that growth may become one of the technology industry’s most important challenges.
