The Vibe Coding Security Crisis:
45% of AI-Generated Code Contains Vulnerabilities

What vibe coding actually means

In February 2025, Andrej Karpathy coined the term "vibe coding": programming where you "fully give in to the vibes, embrace exponentials, and forget that the code even exists."

Within a year, it became the dominant way software gets written. 92% of US developers now use AI coding tools daily. 25% of Y Combinator's Winter 2025 startups had codebases that were 95% or more AI-generated. The productivity gains are real: teams ship faster, prototypes happen in hours instead of weeks, solo developers build things that used to require teams.

The security consequences are also real. As Rekhi wrote on Medium: "We are building the software infrastructure of the next decade on code that nobody understands, maintained by developers who did not write it, secured by defaults that are not secure, scaling at speeds that outpace review capacity."

What the research actually found

Multiple independent studies have measured the security quality of AI-generated code. The findings are consistent across different research teams, different AI models and different programming languages.

What goes wrong in practice

The vulnerability data tells you the scale. The incidents tell you the shape. Here is what happens when vibe-coded applications meet production.

The review gap

AI accelerates code generation by 30-40%. Code review capacity has not increased at all. This is the fundamental asymmetry that creates the security debt.

As Karpathy himself recently warned: as we rely more on AI, our primary job shifts from writing code to reviewing it. But the tooling, the processes and the habits have not made that shift. Developers are still reviewing code at the same pace they did when they were writing it themselves.

The result is what Towards Data Science calls the "security debt crisis": traditional DevSecOps was not built for this volume of code. The pipeline that worked when humans wrote 100% of the code breaks when 41% of it comes from models that optimize for functionality over security.

Why "just use AI to review" does not work

The open source impact

The crisis extends beyond commercial software. Research published in early 2026 argues that vibe coding is killing open source. The argument: when developers stop writing code, they stop understanding the systems they contribute to. Pull requests from vibe coders flood maintainers with code that works but that nobody can explain. The shared knowledge that makes open source ecosystems function erodes.

404 Media reported on the growing burden on maintainers who now spend more time reviewing AI-generated contributions than improving the projects themselves. The code compiles. It passes basic tests. But it introduces patterns that the maintainer did not choose, dependencies that were not needed and security assumptions that were never validated.

Beyond the code: the agent attack surface

Vibe coding's security problems go deeper than vulnerable code output. The tools themselves have become attack targets. The files that control AI behavior, the MCP servers that give AI access to external systems and the agent skill packages that developers share are all surfaces that attackers are actively exploiting.

In March 2026, Mindgard audited 15 major AI coding tools and found 37 security vulnerabilities across Google Gemini CLI, OpenAI Codex, Amazon Kiro, Claude Code, Cursor and others. Every finding led to remote code execution, data exfiltration or sandbox bypasses. These are not theoretical risks. They are confirmed attack chains against products that millions of developers use daily.

What actually works

The vibe coding security crisis will not be solved by asking developers to slow down. They will not. It will not be solved by adding another AI to the review chain. That is circular. It requires tooling that operates at the same speed as AI code generation, catches known vulnerability patterns deterministically and plugs into the workflow developers already use.

The bigger picture

Vibe coding is not going away. The productivity gains are too large. The developer experience is too good. The market pressure to ship fast is too strong. Fighting the trend is not a viable strategy.

But accepting the security debt is not viable either. AI-generated code is already the cause of one in five breaches. The 45% vulnerability rate compounds with the 41% adoption rate. Every quarter, more code ships with more flaws and less review.

The solution is not less AI. It is better guardrails. Deterministic scanning that operates at AI speed. Cryptographic signing that proves agent skills and MCP tools have not been tampered with. Hook-based architecture that intercepts vulnerable code before it reaches the filesystem. The infrastructure exists. The question is whether teams adopt it before their next breach or after.

Further reading