Research shows the vibe coding security crisis CIOs can't ignore

The data that should keep you up at night

In any given year, there is always a good volume of security stats that can keep IT professionals awake at night. Vibe coding introduces another whole set of concerns.

The Red Access research findings

Some of the key findings in the Red Access report include:

• 2,000 of the 5,000 identified vibe-coded applications contained sensitive data with no authentication, no access controls and no audit trail.
• Data types exposed included financial records, patient conversations, strategic documents and credentials and API keys.
• According to Red Access, in a 'meaningful number of cases', applications granted administrative access by default to any visitor who reached the URL.
• Cases included a live financial dashboard at one of Latin America's largest banks and strategic documents from a $200 billion company, reached through a vendor's tool.
• Exposures documented across every industry and every continent, including Fortune 500 enterprises, reached through third-party vendor tools, with organizations, in some cases, passing compliance audits while the exposures remained live.

"What we found was that no sector was clean. Legal, healthcare, government, finance and, with some irony, instances tied to cybersecurity companies, that mix is what reframed it for us," Dor Zvi, CEO of Red Access, said. "This isn't a niche problem in any one vertical. It's a category-level pattern, and it means the work for security teams is both stack work and coaching work because the people building these apps aren't malicious, they're competent employees solving real problems faster than their organization could."

The broader AI coding security landscape

The Red Access report is not the first, nor is it likely to be the last, report on the AI coding risks. The Red Access findings sit inside a broader picture of vibe coding security risks and AI-generated code vulnerabilities.

• Veracode's 2025 GenAI Code Security Report found security weaknesses in 45% of AI-generated code samples.
• Apiiro reported AI-assisted developers shipped code three to four times faster while producing security findings at roughly 10 times the rate.
• A March 2025 study from the University of Texas at San Antonio found nearly one in five AI-generated code samples from the GPT- series referenced software packages that did not exist.
• Escape.tech scanned more than 2,000 critical vulnerabilities across 1,400 production vibe-coded applications, along with exposed secrets and sensitive information.
• Harness reported in its 2025 State of Software Delivery that 67% of developers spend more time debugging AI-generated code than before adopting AI coding tools.

The financial impact

The risk from an insecurely coded application can be material to an organization.

IBM's 2025 Cost of a Data Breach Report provides hard numbers on the costs of shadow AI incidents to organizations.

• Shadow AI incidents average $4.63 million per breach, $670,000 above the baseline for standard breaches.
• 63% of organizations that experienced AI-related breaches lacked an AI governance policy.
• 32% of breached organizations paid regulatory fines, with 48% exceeding $100,000.

Hidden costs often extend further with technical debt from unreviewed AI code, performance remediation and emergency patching as unsecured applications move from prototype to production.

Why this is your biggest blind spot

The challenge for many enterprises is that traditional security controls were designed for IT-controlled application deployment. With vibe coding, that's no longer the case, leading to unsecured AI applications and AI code exposure risks.

The visibility gap. There is a clear visibility gap in the security of AI coding tools, as it resides in an area that traditional IT security doesn't monitor.

"The problem with shadow AI is it skips all of the different types of enterprise controls and processes, which assume that there is a review from governance, procurement, architecture, security and change management," said Erik Avakian, technical counselor at Info-Tech Research Group and former chief information security officer for the Commonwealth of Pennsylvania.

The speed paradox. Vibe coding is fast, enabling users to rapidly go from idea to a publicly reachable production system connected to live enterprise data.

The democratization double-edge. Non-technical employees can now build and deploy full applications, and most have no background in authentication, access controls or data classification.

"These non-technical employees think they're enhancing productivity," said Charles Henderson, head of DivisionHex at Coalfire. "In reality, they're creating a new attack surface that's unmonitored and unsecured.

The "it's just a quick tool" mentality. Employees don't see internal tools as security risks. "It's just for our team" becomes a publicly accessible URL, and prototype becomes production without review because deployment is a single click. "They may share a link without realizing it can be forwarded. They may connect an entire folder when the tool only needs five fields," said Tom Levi, field CISO and director of cyber-risk strategy at CYE, a global cybersecurity company. "That is where a small workflow shortcut can become a material business exposure."

The real-world scenarios

Vibe coding risks are not hypothetical; there are real-world scenarios, including the following examples:

• The sales dashboard. A sales manager builds a deal tracker in a vibe coding tool, connects it to the CRM and publishes without authentication. Customer names, deal values and forecasts sit at a guessable URL, indexed by Google. "The composite version of a case I've seen more than once: a small internal dashboard pulling from a CRM extract, deployed via the platform's default publish to web flow. API key embedded directly in the front-end bundle. No auth gate,"Krti Tallam, senior member of technical staff at KamiwazaAI said.
• The HR tool. An HR coordinator builds an employee directory with salary bands, performance reviews and personal information. The link gets shared internally, and the app is also publicly accessible. The builder never considered regulatory liability or the lack of an audit trail.
• The customer support helper. A business team builds a tool to summarize customer tickets or patient conversations, and security discovers it is processing customer names, contract details and regulated data outside normal controls.
• The financial tracker. A finance team member builds an expense tracker connected to accounting systems. Sensitive financial data is accessible using a public URL, creating a compliance exposure that the builder never considered.

Why traditional security approaches fail

Enterprises have multiple tools and processes in place for security; however, most traditional security approaches fail for several reasons.

The tool proliferation problem

The vibe-coding market spans dozens of platforms, including Lovable, Replit, Base44, Bolt and Cursor, with new entrants constantly. Blocking them all would cut off the work that organizations depend on. "Blocking everything feels decisive, but it usually drives the activity into places security cannot see," Levi said.

The detection challenge

Apps are deployed to Netlify, Vercel and Replit's external hosting, not on corporate infrastructure or networks, generating no signal in traditional asset discovery tools and no DNS records pointing to corporate domains. They appear in no asset inventory. Phishing pages impersonating Bank of America, FedEx, Trader Joe's and Costco on vibe-coding platform subdomains, found by Red Access researchers, inherit the platform's domain reputation, neutralizing domain-based filtering.

"Traditional monitoring was not built to see AI-centric activity," Henderson said. "It misses things like prompt-based data exfiltration, OAuth AI integrations, browser-based tools, model-to-data interactions and API token sprawl."

The immediate action plan

CIOs are not helpless against the risks of vibe coding. There are some immediate steps that can be taken to help manage the risk.

Discovery and assessment

Start with inventory, not policy, and complete it within 30 days. Frame it as an ask, not an audit. "Before any tooling decision, run a workforce-wide inventory ask. Not an audit -- an ask," Zvi said.

Action items:

• Survey development teams on AI coding tool usage.
• Review credit card statements for platform subscriptions and search for corporate email addresses on Lovable, Replit, Base44 and similar platforms.
• Check for apps deployed under corporate accounts on Netlify, Vercel and Replit.
• Identify which teams handle sensitive data, evaluate regulatory compliance impacts and calculate financial exposure.

Immediate risk mitigation

Once you know what exists, act on the highest-risk exposures first. Publicly accessible apps with no authentication are the priority.

Action items:

• • Implement a temporary pause on new vibe-coded app deployments.
  • Require all existing applications to be registered and make clear what happens if teams do not comply.
  • Add authentication to any publicly accessible app immediately and remove sensitive data from apps that cannot be quickly secured.
  • Provide approved platforms with built-in controls as an alternative.

Long-term governance framework

The long-term goal is to enable users to use vibe coding to build secure applications without resorting to a shadow approach.

"Build the paved path," said Will Bengtson, CISO at ConductorOne. "Create the AI skills files, the process, the blessed route for deploying an app inside your organization."

Action items:

• Develop an AI coding governance policy covering acceptable use cases, prohibited data types and approval workflows.
• Deploy discovery tools, extend data loss protection rules to major vibe-coding platform domains and integrate security scanning into approved workflows.
• Provide some training to users within the organization on authentication, access controls and data classification before they build.

"The goal isn't to stop shadow AI coding," Tallam said. "It's to make sure the audit trail survives it."

Conclusion: The choice you face

Vibe coding usage is real and isn't going to disappear. Responsible and proactive CIOs and IT leaders must face the reality shown by the data.

Red Access found 5,000 corporate-purpose vibe-coded applications exposed on the public internet. These apps did not require exploitation to access, and they were not listed in any asset inventory. In other words, this is not a hypothetical risk or a future concern. The apps are live now, and employees are continuing to build more.

For most organizations, the choice to enable AI-assisted development has effectively already been made. The real question is whether CIOs put governance around it before one of these apps becomes the next security incident.

"The honest framing is that they're how you start, not how you finish," Zvi said.

Sean Michael Kerner is an IT consultant, technology enthusiast and tinkerer. He has pulled Token Ring, configured NetWare and been known to compile his own Linux kernel. He consults with industry and media organizations on technology issues.