What 2,000 Exposed Vibe-Coded Apps Reveal About the Limits of Most Security Stacks

# 2,000+ Vibe-Coded Apps Exposed Online: How Shadow AI Bypasses Enterprise Security

The dark side of democratized AI development has fully arrived. A new investigation reveals that employees across six continents are quietly building production applications with AI-powered coding platforms and deploying them directly to the public internet—often granting unrestricted admin access—while enterprise security tools remain blind to the entire operation.

The scale is staggering. Red Access's "Shadow Builders" report, covered in May by Axios, WIRED, and VentureBeat, identified over 380,000 publicly accessible web assets across leading vibe-coding platforms. Of those, approximately 5,000 appeared corporate in nature. More than 2,000 contained sensitive corporate data, operational information, or personal details—exposed without basic access controls, often with default admin privileges open to anyone who discovered the URL.

This represents a fundamentally new class of organizational risk: not malicious insiders or sophisticated attackers, but well-intentioned employees solving real business problems faster than traditional IT could, using tools explicitly designed to make that possible.

## The Threat: Shadow AI at Scale

The exposures span every major industry across six continents. None required exploitation beyond discovering a public URL. Remarkably, many of these exposures existed within organizations that were actively passing security audits and compliance reviews.

The typical scenario:

Each application is then published to the open internet with whatever security controls—often none—the builder configured. The employees aren't reckless. They're competent, working within a platform designed to make this possible, solving problems their organization couldn't solve fast enough through official channels.

Key statistics:

## What is Vibe Coding? A Primer

Vibe coding refers to AI-driven development platforms that allow non-programmers to describe what they want and receive a fully functional application in response. Tools like Bubble, FlutterFlow, Retool, and internal company alternatives have compressed what used to require months of engineering effort into something a competent domain expert can ship before lunch.

The value proposition is real. Speed: A marketing professional can build custom tracking tools without waiting for the development backlog. Autonomy: Business users solve their own problems. Cost efficiency: No engineering resource allocation required.

The platforms deliver exactly what their users asked for. But the platforms were designed for individual projects, small teams, and internal tools—not production-grade applications handling sensitive organizational data and direct integrations with systems of record.

## The Scale of Exposure: 2,000+ Apps and Counting

Red Access's methodology identified nearly 400,000 web assets across the major platforms. Filtering for corporate indicators—domain naming, internal terminology, organizational structure references—narrowed that to roughly 5,000. Of those, more than 2,000 contained immediately sensitive data:

The researchers found no sophisticated attack surface. These weren't hidden vulnerabilities requiring exploit chains or social engineering. They were accessible URLs with default permissions, discoverable through basic search and enumeration. In many cases, the applications granted administrator privileges to anyone who reached the endpoint.

Data categories exposed:

| Type | Frequency |

|------|-----------|

| Business Intelligence / Analytics | 68% |

| Customer or Employee Records | 52% |

| Financial Data (invoices, budgets) | 41% |

| Operational Systems Data | 36% |

| API Keys or Credentials | 18% |

## Why Enterprise Security Stacks Miss This

A Chief Information Security Officer reviewing standard security metrics might reasonably assume the organization is protected:

Yet Shadow Builders slip through every layer.

EDR's blindness: To an endpoint agent, vibe coding looks indistinguishable from ordinary browser activity. A developer building an application inside a browser IDE generates the same telemetry as an employee reading the news—process activity, network calls to known platforms, standard authentication. EDR sees the browser process, not the custom application being built inside it.

DLP's enumerated channels: Data Loss Prevention monitors known exfiltration channels—email, USB drives, known AI chat interfaces. It can flag an employee pasting a customer database into ChatGPT. It cannot see a vibe-coded application programmatically connecting to a sanctioned internal system (like a CRM or BI platform). The connection itself is legitimate; the data flowing through a custom application isn't on DLP's radar.

CASB's application blindness: Cloud Access Security Broker tools monitor access to known SaaS platforms. They see that an employee accessed Retool (the platform). They don't see what application that employee built on it, or whether it's connected to production systems.

Coverage gaps: Personal laptops, contractor machines, BYOD devices, and personal browser tabs exist outside most security perimeters entirely. An employee working from home on a personal laptop can build and deploy a production application without triggering any corporate monitoring.

## Technical Details: How These Apps Connect to Production

The security risk scales dramatically because vibe-coded applications often integrate directly with production systems of record:

The applications are typically hosted on the platform itself (a reputable vendor). The platform infrastructure may be audited and SOC-2 compliant. The custom application built on the platform—and more critically, the data it exposes—is neither audited nor controlled.

## Implications for Organizations

This shift represents a fundamental change in the attack surface. Shadow IT, in the traditional sense, was bounded: when a team bought Trello without IT approval, the data sat in an external SaaS vendor. Audit logs existed. Governance surfaces existed.

Shadow Builders invert that architecture:

A perimeter-focused or identity-based security program will miss this. These applications use legitimate employee credentials to access legitimate platforms. They're accessed from legitimate corporate devices. The risk sits at the intersection of developer autonomy, platform capability, and the gap between where business value is created and where security governance exists.

Who is exposed?

## HackWire Analysis

This isn't a failure of individual security tools. EDR, DLP, and CASB are doing exactly what they were designed to do. This is a failure of governance models to keep pace with the speed of AI-assisted development.

Why this matters now: The Shadow Builders report lands at an inflection point. Vibe-coding platforms have crossed from niche developer tools into mainstream productivity platforms. When the tools were new, adoption was slow and localized. Now they're approved and encouraged as part of "citizen developer" and "democratization" initiatives. The same organizational enthusiasm that drives adoption creates urgency around addressing the exposure.

Pattern recognition: This echoes the BYOD transition fifteen years ago. Organizations resisted for years—"personal devices are less secure"—until the momentum made resistance futile. The reflex was to bolt on MDM and hope for coverage. The actual solution required rethinking identity, trust boundaries, and what "corporate" meant. Vibe coding will follow the same trajectory. Blocking the platforms won't work. The real answer is governance that doesn't require pre-approval for every application, but does require visibility, data classification, and enforced baseline controls.

Hidden risk: The most dangerous exposures aren't the ones found by researchers. They're the ones that haven't been discovered yet. Red Access searched the *public* internet. A Shadow Builder who configured basic authentication, or deployed to a subdomain, or randomized the URL would remain invisible. The 2,000 exposures represent the low-hanging fruit. The actual risk surface is likely multiples larger.

Concrete next step for defenders: Audit your vibe-coding platform usage immediately. Most platforms have admin dashboards showing what applications are deployed. Ask: Who built these? What data do they access? Is it published? Who can access it? Are there API keys embedded in the code? Then implement a simple framework: (1) Classify data the app touches, (2) require a minimum baseline of access controls proportional to the data sensitivity, (3) establish a quarterly audit cadence, (4) don't shut it down—enforce it.

— HackWire Editorial

## Recommendations for Organizations

Immediate actions:

Medium-term controls:

Long-term strategy:

## Related Coverage