Vibe coding might be extremely popular, but it comes with some serious security concerns, including exposed databases, leaked APIs, remotely orchestrated prompt injection attacks, and AI scraping. It's enough to raise the question: Is there a version of vibe coding that could actually hold up in a live environment? I spoke with Shiran Brodie, Head of Growth at Softr, for a face-to-face discussion on the challenges and solutions to vibe coding security hassles.

Softr began as a no-code app builder designed specifically for professionals, small businesses, and enterprises to build apps or portals to support their internal workflows. That means handling a ton of proprietary data, where security is always a big concern. The original version has gone through several updates since launch, with vibe coding now one of the platform's core features.

Brodie spoke about the challenges that the company faced expanding into this new space and the solutions that helped get through them. We also talked about how app development platforms like Softr now suggest a hybrid approach that layers vibe coding features on top of pre-built security infrastructure like built-in database management, access control, and developer visibility. But how does this hybrid security system work on a technical level? And more importantly, does it actually hold up in practice?

Vibe coding is easy, but simplicity makes it a target

In 2025, a security firm called Escape DAST analyzed 5,600 vibe-coded apps built using platforms like Loveable, Replit, Base44, and Bolt. More than 2,000 of these apps had security issues, totaling 2,038 critical vulnerabilities and over 400 exposed credentials. 

“The problem ultimately is that these platforms are generating raw code from scratch,” explained Brodie. Simplicity is a big part of the appeal with vibe coding, but also the thing that can create security issues down the line. “It can abstract a lot of the complexity that comes with building, but non-technical people don't know what they don't know.”

Also in 2025, Georgia Tech researchers launched the Vibe Security Radar, a platform that pulls information directly from publicly accessible security advisories and traces them back to AI-generated coding platforms and LLMs. Researchers scanned more than 40,000 security advisories, and uncovered 43 critical vulnerabilities across eight vibe coding platforms that led to command injections, server-side request forgery, and authentication bypass. Even in the last few months of 2026, security analysts have discovered massive API leaks from Meta-owned Moltbook and the Hugging Face community. Most of these incidents stem from poorly generated code that didn’t take enough steps to secure private API keys from the open web. 

Brodie believes that the problem with vibe coding is that it leaves too much of the core infrastructure up to AI generation, even the mission-critical parts like third-party integrations and permission management systems—which is where most of these vulnerabilities originate.

Contain the AI instead of letting it run free

Instead of letting AI models generate every new app from scratch, Softr retains the same pre-coded backend infrastructure, hosting, integrations, and permission management. Vibe coding features add more design flexibility and smarter workflow automation, but the core systems remain the same.

For example, if you wanted to build an online portal that pulls certain information from your internal company database using an API, Claude Code or Loveable would accomplish this by asking their AI models to write an integration using custom logic, webhooks, and UI wiring. With a platform like Softr, these integrations are available as ready-made components with security best practices already built into them. You can still use vibe coding options to customize their look and functionality, but the core systems and security protocols remain intact.

In development terms, this is called visual scaffolding. Apart from Softr, several other no-code platforms have carried forward this same model, with varying levels of success. Framer, for instance, offers an AI-enabled Workshop extension in its app marketplace. The feature has expanded into Framer’s built-in prompt-based development tool and AI agents that can handle ongoing maintenance tasks. FlutterFlow, a no-code tool designed specifically for building smartphone apps for iOS and Android, also lets you generate pages and components with established security guardrails using AI. 

“It's not an iframe,” Brodie clarified, when describing how vibe-coded elements are isolated from the security layer within Softr. Instead, AI-generated code is confined to specific component blocks that allow custom code to exist with restrictions. “These blocks are built to exist within our infrastructure,” she added.

Dealing with shadow AI and IT sprawl

Shadow AI is the AI-era version of a long-standing problem called "shadow IT." When employees put company data in a software platform without IT approval, it opens businesses to data governance and cybersecurity risks. The same goes for AI, except now, it’s no longer limited to a rogue Trello board, but an employee’s personal account running Claude Code or Gemini with full access to proprietary company databases. AI coding platforms offer generous limits even on personal accounts, making it convenient for employees to go off on their own and start building apps on company data without a centralized approval system. 

“There's no incentive to exist within company guardrails,” Brodie said. Even when employees disclose their vibe-coded apps, IT leaders are expected to maintain and troubleshoot that poorly generated code even though they’d never pass a regular security audit.

Earlier this year, RedAccess scanned more than 380,000 vibe-coded applications built using platforms like Lovable, Replit, or Base44. They found over 5,000 of these apps publicly accessible on the open web and 2,000 of them actively leaking company data, including customers’ sensitive financial and medical information.

It isn’t enough to fix bad coding practices in large language models or vibe coding tools; instead, companies must implement better access controls and improve visibility for system administrators. If your IT team has a bird’s-eye view of who has access to what and the power to control that access, a scenario where rogue apps and portals leak company data becomes a lot less likely.

MCPs make things more complicated

Model Context Protocol—or MCP for short—is Anthropic’s open-source standard for securely connecting AI models to external applications and data sources. It’s quickly become an industry staple, more so because of convenience and the fact that it’s free to use. But the problem with MCP connectors is that they inherit the same permission structure as the database or application it’s connected to. 

Softr has its own database management system that lets you control what MCPs can see. For example, Brodie uses Claude's MCP connector to read and write directly into Softr Databases, but that only stays safe because her databases are governed by the permissions and rules implemented in Softr. Most vibe coding platforms don’t benefit from the same level of access control. 

Vibe coding tools often get blanket access to your raw databases without a permission layer in between, making breaches more likely when attackers use prompt injection attacks. Data is further exposed for companies that need to give third-party vendors and clients access to their vibe-coded applications, where controlling access to company data is a minimum security requirement.

Beyond Softr Databases, the platform offers access to more than 60 pre-coded integrations to platforms like HubSpot, Stripe, and ClickUp. These integrations are already built with security standards in mind, making data leaks less likely. You can also connect Softr to Zapier or Make to benefit from a larger selection of pre-built automations, all of which are more secure than your standard vibe-coded MCP. 

Making security accessible to SMBs, but also enterprises

There’s a marked difference in the security requirements of a small business or early-stage startup versus a large-scale enterprise operation spanning multiple locations. Small business startups are usually content with basic permission management and data encryption standards, whereas enterprise organizations need GDPR and SOC II compliance, control over data center locations, and stronger access control with multi-factor authentication and single sign-on facilities.

Like Softr, more startups are noticing the enterprise security gap posed by vibe coding and trying to build solutions that offer better visibility and governance options. Retool, a low-code app builder, has obtained SOC II Type 1 certification with its strict role-based access control for AI features and limitations to keep LLMs from touching important databases or infrastructure. WaveMaker promises similar security controls, with LDAP-based authentication, centralized permission management, cross-site scripting prevention, and SSO integration through platforms like Okta to woo enterprise clients.

"No organization wants to hire or train more developers just to keep one internal tool secure," Brodie said. Larger companies expect these things to be handled by an AI coding platform’s native security systems. With multiple major public security incidents rattling the AI coding space, no-code platforms are looking for options to offer users more flexibility, without being plagued by the complexity that made them turn to vibe coding in the first place.