Vibe-Encrypted Apps Are a Security Nightmare, but They Can Be Avoided.
Vibe programming may be extremely popular, but it comes with serious security concerns, including vulnerable databases, API leaks, remote hint injection attacks, and AI-powered data mining. This begs the question: is there a version of Vibe programming that can truly withstand real-world testing? I spoke with Sheeran Brody, Head of Development at Softr , to discuss the challenges and solutions surrounding Vibe programming security in person.
Softr began as a no-code app builder designed specifically for professionals, small businesses, and large companies to create apps or portals to support their internal workflows. This means working with vast amounts of sensitive data, where security is always a major concern. The original version has undergone several updates since its launch, and coding with Vibe is now one of the platform’s core features.
Brody discussed the challenges the company faced entering this new market and the solutions that helped overcome them. We also discussed how app development platforms like Softr now offer a hybrid approach that layers Vibe’s coding features on top of a ready-made security framework, such as built-in database management, access control, and developer transparency. But how does this hybrid security system work on a technical level? And more importantly, how effective is it in practice?
Programming on Vibe is simple, but it’s precisely this simplicity that makes it appealing.
In 2025, security firm Escape DAST analyzed 5,600 apps built using frameworks such as Loveable, Replit, Base44, and Bolt. More than 2,000 of these apps had security issues, with a total of 2,038 critical vulnerabilities and over 400 instances of credential leaks.
“Ultimately, the problem is that these platforms generate source code from scratch,” Brody explained. Simplicity is a key part of Vibe Coding’s appeal, but it could also create security issues down the road. “It may abstract away much of the complexity associated with development, but non-technical people don’t know what they don’t know.”
Also in 2025, researchers at the Georgia Institute of Technology launched Vibe Security Radar , a platform that extracts information directly from public security alerts and traces them back to AI-generated coding platforms and LLM platforms. The researchers scanned over 40,000 security alerts and discovered 43 critical vulnerabilities across eight Vibe coding platforms, leading to command injection, server-side request forgery, and authentication bypass. Even in the last few months of 2026, security analysts discovered large-scale API leaks from the Meta-owned Moltbook platform and the Hugging Face community . Most of these incidents were attributed to poorly generated code that failed to take sufficient measures to protect private API keys from public access online.
Brody believes the problem with Vibe-based programming is that it leaves too much of the core infrastructure to the AI generation, even critical parts like third-party integration and permissions management systems—which is where most of these vulnerabilities arise.
Instead of allowing AI to develop unchecked, it should be contained.
Instead of letting AI models create each new app from scratch, Softr maintains the same pre-coded server infrastructure, hosting, integrations, and permissions management. Vibe’s coding features add more design flexibility and more intelligent workflow automation, but the core systems remain the same.
For example, if you want to create an online portal that retrieves specific information from your company’s internal database using an API, Claude Code or Loveable can do this by asking their AI models to write an integration using custom logic, webhooks, and an interface. On the Softr platform, such integrations are available as pre-built components with built-in security guidelines. You can still use Vibe’s coding options to customize their appearance and functionality, but the core security systems and protocols remain unchanged.
In the development context, this is called visual wireframing. Besides Softr, several other no-code platforms have continued this model with varying success. For example, Framer offers an AI-enabled Workshop extension in its app store. This feature has expanded into Framer’s built-in suggestion-based development tool, as well as AI agents that can perform routine maintenance tasks. FlutterFlow , a no-code tool designed specifically for building iOS and Android smartphone apps, also enables AI-powered generation of pages and components with pre-installed security controls.
“It’s not an iframe,” Brody clarified, describing how elements created using Vibe code are isolated from Softr’s security layer. Instead, the AI-generated code is confined to specific component blocks, allowing for the use of custom code with restrictions. “These blocks are designed to work within our infrastructure,” she added.
Combating shadow AI and IT infrastructure sprawl.
Shadow AI is a version of the long-standing problem known as “shadow IT” in the age of artificial intelligence. When employees place corporate data on a software platform without IT approval, it exposes the business to data management and cybersecurity risks. The same applies to AI, except now it’s not limited to an unauthorized Trello board, but to an employee’s personal account using Claude Code or Gemini with full access to the company’s proprietary databases. AI programming platforms offer generous limits even for personal accounts, allowing employees to independently create applications based on corporate data without a centralized approval system.
“There’s no incentive to adhere to corporate rules,” Brody said. Even when employees disclose their applications are written using poor-quality code, IT managers are expected to maintain and troubleshoot this poorly generated code, even if it never passes a routine security audit.
Earlier this year, RedAccess scanned over 380,000 apps built using platforms like Lovable, Replit, and Base44. They found that over 5,000 of these apps were publicly accessible online, and 2,000 of them were actively distributing corporate data, including sensitive financial and medical information.
Simply fixing code errors in large programming language models or Vibe Code tools isn’t enough; instead, companies need to implement more effective access controls and improve transparency for system administrators. When your IT team has a complete understanding of who has access to what and how to manage that access, the likelihood of corporate data leaks from malicious applications and portals is significantly reduced.
MCP modules only complicate the situation.
The Model Context Protocol ( MCP) is Anthropic’s open standard for securely connecting AI models to external applications and data sources. It has quickly become an industry standard, largely due to its ease of use and free nature. However, the challenge with MCP connectors is that they inherit the same permissions structure as the database or application they connect to.
Softr has its own database management system , which allows for control over MCP user access to data. For example, Brody uses Claude’s MCP connector to directly read and write to Softr databases, but this is only secure because her databases are governed by access rights and rules implemented in Softr. Most vibration coding platforms don’t have the same level of access control.
Vibe programming tools often provide unrestricted access to your raw databases without any intermediate permissions, increasing the likelihood of hacks using code injection attacks. Furthermore, this data becomes vulnerable for companies that need to provide third-party vendors and customers access to their Vibe-based applications, where access control to corporate data is a minimum security requirement.
In addition to Softr’s databases, the platform offers access to over 60 ready-made integrations with platforms like HubSpot, Stripe, and ClickUp. These integrations are built with security standards in mind, reducing the risk of data leaks. You can also connect Softr to Zapier or Make to take advantage of a wider selection of ready-made automation scripts, which are more secure than standard MCP scripts created with Vibe.
Making security accessible to small and medium-sized businesses, as well as large enterprises.
The security requirements of a small business or early-stage startup differ significantly from those of a large enterprise operating across multiple locations. For small startups, basic access rights management and data encryption standards are usually sufficient, while larger organizations require GDPR and SOC II compliance, control over data center locations, and stricter access controls through multi-factor authentication and single sign-on.
Like Softr, a growing number of startups are recognizing the gap in enterprise security created by Vibe-based programming and are seeking to build solutions that provide greater visibility and control.Retool , a low-code app builder, has achieved SOC II Type 1 certification thanks to strict role-based access control for AI functions and restrictions preventing low-code users from accessing critical databases or infrastructure. WaveMaker promises similar security controls , including LDAP-based authentication, centralized permissions management, cross-site scripting prevention, and SSO integration through platforms like Okta, to attract enterprise customers.
“No organization wants to hire or train new developers just to secure a single internal tool,” Brody said. Large companies expect these tasks to be addressed by the AI programming platform’s native security systems. With numerous major security incidents rocking the AI programming space, no-code platforms are seeking options that offer users greater flexibility while avoiding the complexities that initially led them to use Vibe programming.