Understanding Vibe Coding | Security Matters — What Organisations Need to Know

• August 17, 2026
• Andrew

Abstract

Vibe coding has delivered real productivity gains across the software industry. It has also introduced a class of security risk that existing governance frameworks were not designed to handle. This article examines the specific security vulnerabilities associated with AI-generated code — the data behind them, the mechanisms that produce them, and the practical measures that organisations can take to capture the speed benefits of AI-assisted development without accumulating security debt that eventually becomes an operational crisis. The evidence is specific and the risks are measurable; this article treats both honestly.

Introduction

Speed is the most frequently cited benefit of AI-assisted software development. Empirical research across Fortune 50 enterprises found that AI-assisted developers produce commits at three to four times the rate of their peers. That productivity gain is real and, for organisations under competitive pressure to ship faster, genuinely valuable.

What the same research found alongside it is less frequently discussed in the same breath. AI-assisted developers introduce security findings at ten times the rate, creating a security debt that accumulates faster than organisations can remediate it.

That ratio — three to four times faster, ten times more security findings — is the central challenge of vibe coding at organisational scale. It is not an argument against AI-assisted development. It is an argument for understanding what the trade-off actually involves and managing it deliberately rather than discovering it during an incident.

The Vulnerability Data

The evidence on AI-generated code security has moved from anecdote to systematic measurement over the past eighteen months. The findings are consistent across multiple independent research programmes.

Veracode tested over 100 large language models on security-sensitive coding tasks and found that 45% of AI-generated code samples introduce OWASP Top 10 vulnerabilities — a pass rate that has not improved across multiple testing cycles from 2025 through early 2026 despite vendor claims to the contrary. OWASP Top 10 vulnerabilities are not exotic edge cases. They are the most well-documented, most widely understood, and most commonly exploited classes of software security flaw in existence. SQL injection, cross-site scripting, broken authentication — the vulnerabilities that security training has been teaching developers to avoid for two decades are appearing at scale in AI-generated code because the models were trained on historical code that contains them.

Veracode found cross-site scripting vulnerabilities in 86% of AI-generated code samples tested across five major large language models. AI-assisted commits expose secrets at twice the rate of human-written code — 3.2% versus 1.5%.

GitGuardian’s State of Secrets Sprawl 2026 report documented 28.65 million new hardcoded secrets in public GitHub commits during 2025 — a 34% year-over-year increase representing the largest single-year jump ever recorded. AI-assisted commits showed a 3.2% secret-leak rate compared to a 1.5% baseline, indicating that AI-generated code roughly doubles the baseline credential exposure rate.

API security firm Escape.tech scanned over 1,400 vibe-coded production applications and found that 65% had security issues and 58% contained at least one critical vulnerability, including over 400 exposed secrets and 175 instances of exposed personally identifiable information such as bank account data.

Why the Vulnerabilities Happen

Understanding why AI-generated code introduces these vulnerabilities at elevated rates matters, because it points directly to what mitigation actually requires.

Because LLMs generate code by reproducing statistical patterns from public repositories, they can also reproduce insecure approaches found in their training data. The issue is not that AI writes insecure code by default. The issue is that AI can produce convincing code faster than teams can reason about it.

AI coding tools can reproduce insecure patterns from training data or generate flawed logic under pressure to produce fast results. That can lead to injection flaws, weak authentication, broken authorisation, or unsafe handling of sensitive data. AI-generated code often introduces open-source packages, frameworks, and libraries automatically. Without validation, teams can inherit vulnerable, malicious, or simply inappropriate dependencies that expand software supply chain risk.

There is also a subtler problem that compounds over time. Security degrades across repeated follow-up prompts. After five iterations of vibe-based refinement, code can contain 37% more critical vulnerabilities than the initial version. Iterating on AI output does not self-correct security flaws. In many cases it compounds them, because each additional prompt introduces new dependencies, new assumptions, and new surface area that the previous iterations did not account for.

One of the most subtle risks is the psychological distance between the developer and the generated code. When a developer “vibes” an application into existence rather than typing it line by line, they may feel less authorship over the result. This ownership paradox weakens accountability — if the AI appeared to author the logic, the human operator is less likely to feel the author’s burden to defend that code against edge cases or deep-seated vulnerabilities.

Tracking the Damage

In May 2025, Georgia Tech’s Systems Software and Security Lab launched the Vibe Security Radar project to track how many publicly filed CVEs can be traced to AI-generated code. The trend line the project has documented is notable. In January 2026, six CVEs were attributable to AI-generated code. In February, fifteen. In March 2026, thirty-five CVEs were directly attributable to AI coding tools. The researchers estimate the true count across the open-source ecosystem is between five and ten times higher, because most AI coding tools do not leave identifiable commit metadata.

The attack surface has also expanded in new directions that did not exist in traditional software development. In April 2026, a design-level command injection vulnerability was disclosed across multiple AI coding SDK implementations, affecting roughly 200,000 vulnerable instances. Cursor, Windsurf, and GitHub Copilot were among the tools affected through their configuration interfaces. These are vulnerabilities in the infrastructure that connects AI coding assistants to the systems they operate on — a category of risk that did not exist at scale eighteen months ago.

What Responsible Organisations Are Doing

The response across organisations that have thought about these findings carefully does not involve abandoning AI-assisted development. It involves treating AI-generated code with appropriate scepticism rather than default trust.

In the 2026 software landscape, productivity should not be measured only by the volume of code generated. A more useful measure is the Recheck-to-Code Ratio. If vibe coding saves a developer four hours of manual syntax work, that time should be reinvested into security design, verification, and logic review. When saved time is used only to ship more features at higher speed, the organisation accumulates security debt.

The practical governance framework that experienced security teams recommend combines several layers. Mandatory code review for all AI-generated output — not as a formality but as a substantive check of logic, authentication flows, and edge cases — is the non-negotiable foundation. Automated security scanning in the continuous integration pipeline catches known vulnerability classes. Secrets management through environment variables rather than hardcoding removes the most immediately exploitable credential exposure vector.

Beyond tooling, the more fundamental shift is cultural. The speed advantage of vibe coding becomes a liability when it trains teams to skip verification steps that matter in production. Production software is not just code that runs — it is code that handles real data, enforces permissions, maintains audit logs, and operates within compliance boundaries.

The EU Regulatory Dimension

For organisations operating within the EU, the security risks of vibe coding intersect directly with a regulatory environment that is already in active enforcement. The EU AI Act (Regulation EU 2024/1689) entered into force on 1 August 2024 and is being applied in phases. Prohibited AI practices and AI literacy obligations have been applicable since February 2025. Transparency obligations for AI-generated content started to apply on 2 August 2026. Where AI systems also process personal data, GDPR obligations apply concurrently — both regimes can apply to the same system, and running them as separate compliance workstreams introduces duplication and gaps. Integrated governance is the more efficient and more defensible approach.

The practical implication for any organisation deploying vibe-coded applications is specific. Whether AI systems, such a standard AI coding assistants used to help developers write code trigger the AI Act’s Annex III high-risk classification, depends on the use-cases in question. As examples, AI systems used for recruitment or selection of workers, termination or promotion of staff, or credit scoring of persons could be classified as high-risk AI systems under the EU AI Act. A vibe-coded application that incorporates any of these functions, or that processes personal data at scale, is not exempt from the compliance requirements which may apply to the same applications built by conventional means, pursuant to the respective legislative frameworks. A security failure in an AI-generated application that exposes personal data leading to a data breach is a GDPR incident. One would also have to assess whether any infringements of other legislations would also have taken place. The method of code generation does not change the obligation or the liability.

The EU’s July 2026 Action Plan on Cybersecurity and AI has further formalised the connection between AI-generated code and cybersecurity governance, setting out a coordinated approach to help member states, businesses, and public authorities address the cybersecurity and resilience challenges posed by AI systems at scale. OWASP’s 2025 Top 10 release and the wider application security community have increasingly highlighted risks associated with AI-assisted software development, including concerns around insecure code generation, software supply chains, and the governance challenges created by rapidly accelerating development workflows. Combined with active EU enforcement activity, this signals that AI-generated code security is no longer a niche engineering concern. It is increasingly a mainstream governance challenge with technical, operational, and regulatory implications.

References

1. Iterasec, Vibe Coding Security: Risks, Vulnerabilities, Best Practices, June 2026. https://iterasec.com/blog/ai-assisted-software-development-security-risks/
2. Cloud Security Alliance, Vibe Coding’s Security Debt: The AI-Generated CVE Surge, April 2026. https://labs.cloudsecurityalliance.org/research/csa-research-note-ai-generated-code-vulnerability-surge-2026/
3. Cloud Security Alliance, AI-Generated Code Security — Vibe Coding 2026, March 2026. https://labs.cloudsecurityalliance.org/research/csa-research-note-ai-generated-code-security-vibe-coding-202/
4. BeyondScale, Vibe Coding Security Risks: Enterprise Guide 2026, April 2026. https://beyondscale.tech/blog/vibe-coding-security-risks-enterprise
5. Checkmarx, Vibe Coding Security: Risks, Vulnerabilities, and Secure AI Coding, April 2026. https://checkmarx.com/blog/security-in-vibe-coding/
6. IBM, Vibe Coding Security Risks Aren’t Like Ordinary Security Risks, June 2026. https://www.ibm.com/think/insights/vibe-coding-security-risks
7. Retool, The Risks of Vibe Coding: Security Vulnerabilities and Enterprise Pitfalls, May 2026. https://retool.com/blog/vibe-coding-risks
8. Sneh Bavarva / Medium, Secure Vibe Coding in 2026: The Files, Prompts and Rules of Use and Research, June 2026. https://snehbavarva.medium.com/secure-vibe-coding-in-2026-the-files-prompts-and-rules-of-use-and-research-e821021ee908
9. Superblocks, Vibe Coding Security: 7 Risks and How to Fix Them in 2026, June 2026. https://www.superblocks.com/blog/vibe-coding-security
10. Modall, Vibe Coding Security Risks: What Founders Need to Know (2026), April 2026. https://modall.ca/blog/vibe-coding-security-risks
11. GitGuardian, State of Secrets Sprawl 2026, March 2026. https://www.gitguardian.com/state-of-secrets-sprawl-report-2026
12. Keyhole Software, Vibe Coding Trends 2026, June 2026. https://keyholesoftware.com/vibe-coding-trends-2026/
13. Augment Code, The 2026 EU AI Act and AI-Generated Code: What Changes for Dev Teams, June 2026. https://www.augmentcode.com/guides/eu-ai-act-2026
14. Legiscope, EU AI Act Practical Compliance Guide 2026, July 2026. https://www.legiscope.com/blog/eu-ai-act-compliance-guide.html
15. European Commission, AI Act — Regulatory Framework and Enforcement Timeline, July 2026. https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai

 This article is based on information gathered from various online sources and intended only for informational purposes. It does not constitute professional advice or recommendations. While the information provided is accurate and true to the best of our knowledge, we make no representations or warranties of any kind, express or implied, regarding the completeness, accuracy, reliability, suitability, or availability of the topics covered. We strongly advise readers to conduct further research and consult appropriate professionals or experts before making any decisions based on the information presented in this article.

We disclaim any liability for losses or damages incurred based on the information provided herein.

This document was prepared with the assistance of generative AI tools for editorial support and image generation. Any AI-generated suggestions were reviewed, edited, and validated by the authors, who remain fully responsible for the final content.