2,000 Exposed Vibe-Coded Apps Highlight Critical Gaps in Security Infrastructure

The concept of Shadow AI has evolved significantly. Initially, it referred to employees inadvertently sharing sensitive information with AI platforms like ChatGPT. Today, it encompasses a broader and more alarming trend: employees creating full-fledged applications using AI tools, integrating them into production systems, and making them publicly accessible—all without the oversight of security or IT departments.

This shift from simple prompts to complex products has expanded the risk landscape considerably.

In the report titled The Shadow Builders, Red Access conducted a comprehensive investigation that revealed over 380,000 publicly accessible web assets across leading vibe-coding platforms. This report, covered by major outlets such as Axios, WIRED, and VentureBeat, highlights a critical issue in cybersecurity.

Among these assets, approximately 5,000 appeared to be corporate in nature. Alarmingly, more than 2,000 of these contained sensitive corporate, operational, or personal data exposed on the open web. These applications were often deployed without basic access controls, frequently granting administrative access to anyone who accessed the URL. This issue spans six continents and affects every industry, with no exploitation required for access.

Organizations are passing audits while these vulnerabilities remain active.

The New Shadow AI: From Prompts to Products

Vibe coding represents a new frontier in AI-driven development platforms, allowing individuals to create functional applications by simply describing their needs. This capability has drastically reduced the time required for development, enabling non-developers to launch applications in a fraction of the time it would take traditional engineering teams.

For instance, a marketing manager might create a campaign tracker linked to a business intelligence (BI) tool, while an operations manager could develop a vendor intake form connected to a ticketing system. Finance teams are also leveraging these tools to create dashboards that pull invoice data, often publishing these applications to the open internet with minimal or no access controls.

The individuals engaging in this practice are not acting maliciously; they are competent employees addressing real challenges more swiftly than their organizations can. The platforms facilitating this development are not inherently problematic; they are responding to user demand. However, the necessary guardrails—both technical and behavioral—have not kept pace with this rapid evolution.

This situation differs significantly from traditional Shadow IT. In the past, Shadow IT was limited in scope; for example, a team might purchase a Trello account without notifying anyone, but the data remained within a known, unsanctioned SaaS vendor. In contrast, Shadow Builders create custom applications with direct connections to production systems, often publishing them on the open internet. While the underlying platform may be subject to audits, the applications built upon it typically are not. The builder, the platform, and the URL exist in isolation from IT oversight.

Why a Mature Security Stack Still Misses This

When faced with the alarming statistics from the report, a Chief Information Security Officer (CISO) might instinctively check their security stack. While tools such as Endpoint Detection and Response (EDR), Data Loss Prevention (DLP), Cloud Access Security Brokers (CASB), firewalls, and Secure Service Edge (SSE) are operational, they often fail to address the gaps created by the emergence of Shadow AI.

EDR solutions monitor browser processes but do not capture the nuances of application builds occurring within those processes. To an endpoint agent, a user leveraging a vibe-coding platform appears to be engaging in ordinary, non-malicious browser activity, akin to reading news articles. Furthermore, modern EDR tools and enterprise browsers only monitor devices owned by the organization and browsers they manage, leaving personal laptops and contractor devices invisible.

DLP solutions focus on specific channels and can flag regulated data being pasted into known AI chats. However, they cannot detect a vibe-coded application that connects to a sanctioned BI tool via API, facilitating data transfers that bypass the endpoint entirely.

CASB tools were designed to manage Shadow IT involving SaaS vendors with identifiable users. They struggle to differentiate between a vast array of custom applications hosted on vibe-coding platforms and the platforms themselves, often categorizing them as a single approved vendor.

Firewalls and SSE can see traffic directed to the platform’s domain but lack the context of applications as business objects. Many SASE/SSE deployments are incomplete, leaving unmanaged device issues unresolved.

These tools are not failing; rather, the new category of risk exists in the gaps between existing security architectures, generating fragmented signals that do not form a cohesive, governable picture.

Where Visibility Actually Has to Live

Vibe coding is fundamentally a web-session event. Each step—from building an application to granting OAuth permissions that link it to sanctioned enterprise systems—occurs within the session layer. The data utilized by the application moves through this session, and the deployment action that publishes the application to a public URL is executed in the same browser tab.

To effectively monitor this activity, controls must be positioned at the session layer. This approach allows for comprehensive visibility into the entire build path, including the platform used, the corporate systems connected, the data being transferred, and the publication event that places the application online. This visibility is attributable to specific individuals and application instances, regardless of the browser or network path used, and is independent of whether the device is corporate-issued or personal.

What to Do This Week

Organizations can take several steps to address this emerging risk without making immediate technology purchases.

First, initiate a discovery phase by directly asking employees what they have built. Most Shadow Builders are engaged in productive work and are not attempting to conceal their activities. A workforce-wide prompt—encouraging employees to disclose any tools created using AI development platforms—can yield more information than a formal policy memo or tool deployment.

Next, map the applications that are surfaced. For each application, identify which corporate systems it connects to, the method of connection (OAuth, API key, manual upload), and whether it is publicly accessible. Public reachability is a critical signal that can be acted upon in the short term.

Establish a sanctioned path for Shadow Builders to report their creations. Clearly define approved platforms, acceptable data categories, and minimum authentication standards to facilitate compliance.

Finally, recognize that this is not a one-time inventory. Vibe-coded applications will continue to emerge, and the landscape will evolve. A mature approach requires continuous discovery at the layer where these activities occur.

The category of Shadow AI will continue to develop, and platforms will adjust their defaults in response. However, the vulnerabilities identified in many enterprises today remain pressing.

Red Access offers an agentless, session-layer security platform designed to provide visibility and governance across any browser and device, including unmanaged ones. This solution can be deployed rapidly, enhancing an organization’s security posture.

Source: thehackernews.com

Keep reading for the latest cybersecurity developments, threat intelligence and breaking updates from across the Middle East.