Vibe Coding Security Crisis: 380,000 Apps Exposed
While everyone celebrates how AI coding tools let anyone build apps in minutes, a sobering reality just emerged: most of those apps are leaking your data to the entire internet. Israeli cybersecurity firm RedAccess published findings on May 7, 2026 that should make every developer pause before their next vibe coding session.
The numbers are staggering. RedAccess found approximately 380,000 applications built with AI coding tools that were publicly accessible on the open web. Of those, about 5,000 contained sensitive corporate and personal data, including medical records, financial information, and internal business documents. This is not a theoretical risk. This is data actively indexed by Google and discoverable by anyone with a search query.
What Got Exposed
| Category | Examples Found |
|---|---|
| Healthcare | Patient conversations at children’s care facilities, clinical trial statuses, doctor-patient summaries |
| Financial | Brazilian bank internal financials, customer payment data |
| Corporate | Shipping schedules, incident response docs, customer service transcripts |
| Educational | School class recordings, student information, teacher schedules |
The exposed applications included a shipping company revealing vessel port schedules, a British healthcare company’s clinical trial statuses, customer service chat transcripts from a furniture company, and internal financial data from a Brazilian bank. Researchers also found phishing sites impersonating Bank of America, FedEx, Trader Joe’s, and McDonald’s built entirely with Lovable.
Why This Keeps Happening
The platforms involved include Lovable, Base44 (owned by Wix), Replit, and Netlify. The fundamental problem is straightforward: privacy settings on many vibe coding tools default to public unless users explicitly change them. When you prompt an AI to build your app and hit deploy, that app often becomes publicly accessible immediately.
This creates a perfect storm for data exposure. Non-technical users building apps do not understand deployment implications. Default settings favor convenience over security. Many applications are indexed by search engines within hours. And there is no security review before code goes live.
Replit’s CEO stated that “public apps being accessible on the internet is expected behavior” and that “privacy settings can be changed at any time with a single click.” While technically accurate, this response misses the point entirely. Users who do not understand the security implications of their choices cannot make informed decisions about those settings.
The Broader Pattern
This incident is part of a larger trend in AI coding tool security that should concern every developer. A large-scale scan by Escape.tech of 5,600 publicly deployed vibe-coded applications found 2,000 highly critical vulnerabilities, 400 exposed secrets including API keys and access tokens, and 175 instances of personally identifiable information.
Georgetown CSET found XSS vulnerabilities in 86% of AI-generated code samples tested across five major LLMs. The Cloud Security Alliance’s 2026 report shows that AI-assisted commits expose secrets at twice the rate of human-written code: 3.2% versus 1.5%. In a December 2025 study testing five major AI coding agents, every single one introduced SSRF vulnerabilities in the same type of feature.
The democratization of software development is real and valuable. But democratizing creation without democratizing security knowledge creates systemic risk that extends far beyond individual applications.
Why AI Engineers Should Care
Through implementing production AI systems across multiple organizations, I have observed a consistent pattern: the speed gains from AI coding tools often come at the expense of security fundamentals that traditional development processes enforce.
Professional engineering teams have code review gates, security scanning pipelines, staging environments, and deployment approvals. Vibe coding bypasses all of these. The same features that make these tools accessible to beginners also remove the guardrails that protect production systems.
This matters for AI engineers specifically because:
Reputation risk scales with capability. As AI tools become more powerful, the applications they produce become more sophisticated. More sophisticated applications handle more sensitive data. When those applications leak, the damage compounds.
Enterprise adoption depends on trust. Organizations evaluating AI coding tools will factor in incidents like this. The Five Eyes agentic AI security guidance already emphasizes treating AI systems with existing security frameworks. These exposure incidents reinforce that message.
You will inherit this code. Applications built quickly with vibe coding tools often need professional maintenance. AI engineers increasingly find themselves auditing and securing applications that were built without security considerations.
Practical Protections
If you are using AI coding tools, whether Claude Code, Cursor, or any vibe coding platform, implementing basic security hygiene is not optional.
Before deployment, verify privacy settings. Do not assume defaults are secure. Check every platform’s deployment configuration explicitly. If the platform does not make security settings obvious, that itself is a warning sign.
Treat AI-generated code as untrusted input. Run security scans on generated code before deployment. Look specifically for exposed secrets, injection vulnerabilities, and improper access controls. The statistics on AI-generated vulnerabilities are clear: assume the code has problems until proven otherwise.
Separate development from production. Use staging environments even for small projects. This creates a natural checkpoint where you can review what is actually being deployed before it reaches the public internet.
Audit existing deployments. If you have built applications with these tools in the past, check them now. RedAccess found applications containing data that had been exposed for months without the creators realizing it.
Organizations that implement governance layers for AI-generated code have seen measurable results. ISACA documented a 36% reduction in remediation time in a 2026 framework study without meaningful reduction in developer velocity. Security and speed are not inherently in conflict if you build the right processes.
The Real Question
The vibe coding revolution is not going away. These tools provide genuine value, and the ability to build functional applications from natural language prompts represents a real shift in how software gets created. But the security implications cannot be an afterthought.
The companies building these platforms have a responsibility to make secure defaults the norm, not an option buried in settings. Until they do, every developer using these tools carries the burden of understanding implications that the tools themselves obscure.
For those serious about deploying AI applications in production contexts, this incident serves as a reminder: the gap between generating code and shipping secure software is exactly where professional engineering expertise provides value. Speed without security is just accelerated failure.
Recommended Reading
- AI Coding Tools Are Under Attack: Developer Security Guide
- Five Eyes Agentic AI Security Guidance for Engineers
- Production AI Systems Development
Sources
To see how to build AI systems with proper security foundations, watch the implementation tutorials on YouTube.
If you want to develop production-ready AI engineering skills with security built in from day one, join the AI Engineering community where members follow 25+ hours of exclusive AI courses, get weekly live coaching, and work toward $200K+ AI careers.
