Updated May 2026

Vibe coding security statistics

The numbers behind the security posture of vibe-coded applications — drawn from peer-reviewed research, verified CVE disclosures, and primary-source publications. Every data point on this page has a direct link to the original source. Cite freely.

Industry-wide research

89.5%

AI-built apps with vulnerabilities

Peer-reviewed audit of vibe-coded applications: only 10.5% of apps in the study were secure.

Source: SusVibes (arXiv:2512.03262) · Dec 2025

98%

Apps missing basic security controls

Tenzai's audit of popular AI coding agents found 98% of resulting apps shipped without basic security controls.

Source: Tenzai · 2025

175

PII records exposed in a single audit

Escape Security's methodology piece documented 175 personally identifiable records exposed across vibe-coded apps they scanned.

Source: Escape Security · 2025

CVE-2025-48757

Lovable RLS bypass disclosed May 2025

Affected 10.3% of analyzed Lovable apps — 303 endpoints across 170 projects out of 1,645. Allowed unauthenticated reads of arbitrary tables.

Source: Matt Palmer (statement on CVE-2025-48757) · May 2025

40%

Stanford: AI-assisted code with vulnerabilities

Stanford controlled study found participants using AI coding assistants produced code with security vulnerabilities ~40% of the time on security-sensitive tasks — and believed the code was more secure than non-AI code.

Source: Stanford University (arXiv:2211.03622) · 2023

Statistics by platform

Verified funding, usage, and security data for each major vibe-coding platform. One page per platform — every claim linked.

Primary research sources

Methodology & sourcing

Every figure on this page comes from a publicly-citable source with a direct link. We exclude any statistic we can't trace back to a primary publication — even widely-repeated numbers — because if a source can't be verified, it shouldn't be cited.

SusVibes (arXiv:2512.03262): Peer-reviewed audit study of vibe-coded applications. Sample size and methodology are documented in the paper.

Tenzai: Comparative audit of popular AI coding agents including Lovable, Bolt.new, Cursor, Replit, v0, and Claude. Methodology described in the linked write-up.

Escape Security: Live application scanning of deployed vibe-coded apps. Methodology piece linked above.

Stanford (arXiv:2211.03622): Controlled experiment with developers performing security-sensitive coding tasks with and without AI assistance.

CVE-2025-48757: Coordinated disclosure by Matt Palmer affecting Lovable apps using Supabase RLS.

Scan your app

VAS scans your live vibe-coded app for the issues this research describes — exposed keys, missing RLS, broken auth — and gives you copy-paste fixes for your AI tool.