In many cases, public access remained enabled long after deployment, creating silent data leaks that went unnoticed for months.
Many of the vulnerable apps were created using platforms like Replit, Netlify, Base44 owned by Wix, and Lovable. Nearly 2,000 apps appeared to contain genuine sensitive information, including advertising spending reports, company strategy documents, chatbot logs, customer contact details, hospital personnel records, and financial summaries.
According to RedAccess researcher Dor Zvi, the issue is linked to the rise of “vibe coding,” where non-technical employees use AI tools to rapidly build and publish web applications. Since these platforms make development extremely simple, apps can go live within minutes without any review from engineering or cybersecurity teams.
Researchers found the exposed apps through basic Google and Bing searches because many AI coding services host projects publicly on shared domains by default.
Some applications exposed private information without requiring logins, while others reportedly allowed outsiders to gain administrative control over backend systems.
The exposed data covered multiple industries. Hospital staff schedules listing doctors’ identities appeared alongside marketing strategy presentations, shipping records, retailer chatbot conversations, and detailed advertising campaign budgets. Such leaks could expose sensitive competitive information, including business planning timelines and financial allocations.
The investigation also uncovered phishing websites hosted directly on AI coding platform domains. These fake pages impersonated major companies including Bank of America, Costco, FedEx, Trader Joe’s, and McDonald’s.
The platforms disputed parts of the findings while acknowledging that publicly accessible apps existed. Amjad Masad said users choose whether apps remain public or private. Lovable emphasized that creators are responsible for configuring security correctly, while Wix stated weakening protections requires deliberate user actions.
Security experts argue the broader issue remains serious because AI coding tools rarely enforce strong safeguards automatically. Many employees using them lack training in authentication systems or permission controls, allowing insecure deployments to slip through unnoticed.
Researchers say the situation resembles earlier waves of exposed Amazon S3 cloud storage buckets, where confusing defaults and user mistakes left sensitive files publicly accessible.
AI-powered coding platforms may now be accelerating similar risks on a larger scale as businesses increasingly rely on AI tools for internal dashboards, marketing systems, client portals, and reporting applications.
Experts also warn the true scale may be far larger. The 5,000 discovered apps only included projects hosted directly on AI platform domains. Thousands more could exist on privately owned domains that standard searches cannot easily detect.
As AI-generated development grows rapidly, companies are now under pressure to strengthen oversight, improve employee training, and introduce stricter security reviews. Without stronger safeguards, fast AI-assisted app creation could continue exposing confidential corporate and personal information online.
