Data Report
State of Vibe Coding Security
2026
The most comprehensive collection of vibe coding security data available. Two documented data breaches. TeamPCP supply chain attack ESCALATING: LiteLLM (Mar 24), Telnyx (Mar 27), now partnered with Vect Ransomware Group (Mar 28) -- 300K dark web users accessing ransomware tools targeting affected organizations. NCSC government warning at RSA Conference. 19+ security scanners launched. Aggregated from Escape.tech, Tenzai, CodeRabbit, Kaspersky, Veracode, Wiz, Sonar, Tenable, The Record, IT Pro, The Cyber Express, DreamFactory, Endor Labs, and independent research.
Last updated: April 4, 2026 (12:35 AEST) | Sources linked below
The Numbers
5,600
Apps Scanned
Source: Escape.tech
Publicly deployed vibe-coded apps built with Lovable, Bolt.new, Base44, and similar tools
2,000+
Vulnerabilities Found
Source: Escape.tech
Security vulnerabilities across scanned applications
400
Exposed Secrets
Source: Escape.tech
API keys, database credentials, and tokens committed to public repos or exposed in production
69
Vulns in 15 Apps
Source: Tenzai
Vulnerabilities found across 15 apps built with 5 AI coding tools, including critical SSRF and injection flaws
2.74x
More XSS Vulns
Source: CodeRabbit
AI co-written code introduces 2.74x more cross-site scripting vulnerabilities than human-only code
45%
AI Code Has Vulns
Source: Kaspersky
Of AI-generated code contains security vulnerabilities
42%
Code Is AI-Generated
Source: Sonar
Of all code written in 2025-2026 is AI-generated
10.3%
Critical RLS Flaws
Source: Lovable Internal
Of Lovable-built apps had critical Row Level Security misconfigurations before Lovable 2.0 built-in scanning
200K
New Projects Daily
Source: TechCrunch / Lovable
New vibe-coded projects created on Lovable alone every day
0/15
Had Security Headers
Source: Tenzai
Zero out of fifteen tested apps had proper security headers (CSP, HSTS, X-Frame-Options, CORS)
1/15
Had Rate Limiting
Source: Tenzai
Only one out of fifteen apps had rate limiting, and it was bypassable
60%
Failed Our Audit
Source: VibeCheck / notelon.ai
Of randomly selected Lovable-built repos on GitHub failed the VibeCheck security scan with grade D or F
Hours
To First Breach
Source: Baudr / Grenbaud
Baudr social network built with AI for 40 euros was hacked within hours of launch. Admin panel open at /admin. Thousands of users' data exposed.
443
Malicious AI Files
Source: McAfee Labs
Malicious zip files using vibe-coded malware detected in January 2026, impersonating AI tools, game mods, and utilities
1.5M
Auth Tokens Exposed
Source: Wiz / Moltbook
Social network Moltbook, built entirely via vibe coding, had misconfigured database exposing 1.5 million authentication tokens and 35,000 email addresses to the public internet
67%
Critical Vulns (100 Repos)
Source: ShipSafe
ShipSafe scanned 100 AI-generated repositories: 67% had critical vulnerabilities, 45% hardcoded secrets, 38% missing auth on API routes, 89% of Lovable apps missing RLS
318
Vulns in 100 Apps
Source: VibeWrench
VibeWrench scanned 100 public GitHub repos built with Lovable, Bolt.new, Cursor, v0.dev: 65% had security issues, 58% had critical vulns, 70% missing CSRF, 41% exposed secrets, 21% no auth on endpoints
81%
Devs Using AI
Source: CodeSignal / Tenable
81% of developers surveyed are using AI for development. Some large tech companies now mandate AI use for developers. Tenable warns citizen developers are deploying AI code without any security checks.
322%
More Privilege Escalation
Source: Apiiro
AI-assisted code introduces 322% more privilege escalation paths compared to human-only code. By June 2025, AI-generated code was introducing over 10,000 new security findings per month -- a 10x spike in six months.
62%
AI Apps Have Vulns
Source: Cloud Security Alliance
62% of AI-generated applications contain design flaws or known security vulnerabilities, even with the latest AI models. Root problem: AI does not understand your application's risk model or threat landscape.
97%
Devs Use AI Tools
Source: GitHub Survey
97% of developers now use AI coding tools, most without any security scanning. 6.4% of repos using GitHub Copilot leak at least one secret -- 40% higher than repos without AI assistance.
1 in 5
CISOs Hit by AI Code
Source: Aikido Security
One in five CISOs surveyed had suffered major incidents because of AI-generated code. 69% of security leaders, security engineers, and developers across Europe and the US found serious vulnerabilities in AI-written code.
1,000+
Enterprise SaaS Hit
Source: Arctic Wolf
At least 1,000 enterprise SaaS environments may be affected by the TeamPCP supply chain attack campaign targeting Trivy, Checkmarx, and LiteLLM. Downstream impact potentially extends to additional projects.
35
CVEs in March Alone
Source: Georgia Tech / Infosecurity Magazine
35 new CVE entries disclosed in March 2026 directly caused by AI-generated code, up from 6 in January and 15 in February. Tracked by Georgia Tech SSLab Vibe Security Radar across 50 AI coding tools including Claude Code, Copilot, Cursor, Devin, Windsurf.
74
AI CVEs Total
Source: Georgia Tech SSLab / The Register
74 CVEs attributable to AI-authored code out of 43,849 advisories analyzed (May 2025-Mar 2026). Claude Code: 49 (11 critical). GitHub Copilot: 15 (2 critical). Real number likely 5-10x higher due to detection blind spots. Researcher: 'We do not think AI code is orders of magnitude safer than human code.'
4%
GitHub Commits Are AI
Source: Claude's Code / The Register
Claude Code alone appears in more than 4% of all public commits on GitHub. Over 15 million total commits. 30.7 billion lines of code added by Claude Code in the past 90 days.
4 TB
Stolen From $10B Startup
Source: SecurityWeek / Fortune / TechCrunch
Mercor, a $10 billion AI recruiting startup, confirmed breached via the LiteLLM supply chain attack. Lapsus$ claims 4 terabytes of data stolen including source code, databases, VPN accounts, Slack data, and contractor information. First confirmed major corporate victim of the TeamPCP campaign.
The Headline
Vibe coding is the fastest way to build software in 2026. It is also producing the largest wave of insecure applications in the history of software development. The data is no longer theoretical.
5,600
apps scanned by Escape.tech in the largest vibe coding security study ever conducted
2,000+
vulnerabilities found across those apps, including exposed secrets, missing auth, and open databases
0/15
apps in the Tenzai study had proper security headers. None. Not one. Across all 5 AI coding tools.
10 Vulnerability Types
6 Platform Breakdowns
2 Breach Case Studies
24-Event Timeline
Check If Your App Is Vulnerable
VibeCheck scans your GitHub repo or live site for the exact vulnerabilities described in this report. Free. No signup. Results in seconds.
Worried About Your App?
Get a professional security audit of your vibe-coded app. 50+ checks, actionable fix prompts you can paste directly into Lovable/Cursor/Bolt. Results in 24 hours.
Get Your Audit - $99This report is maintained by notelon.ai and updated as new data becomes available.
Cite as: "State of Vibe Coding Security 2026, notelon.ai/report"
Questions or additional data? @solobillionsHQ on X