Not verified
First-quarter 2026 assessment of 200 apps
183 of 200 vibe-coded apps, 91.5 percent, contained at least one vulnerability traceable to AI hallucination or missing security context.
2026-04Mixed
Incidents and data
The launch failure library
The launch failure library is a sourced record of what has gone wrong in AI-built apps. Every entry maps the reported failure to what an outside check could have observed, or states plainly why it could not.
16 incidentsLast updated September 13, 2026
Root cause
Covered by status
- Declared
CVE growth in AI-generated code
CVE entries attributed to AI-generated code rose from 6 in January 2026 to more than 35 in March 2026.
2026-03Mixed - Could not check
Mercor supply-chain breach
A 10 billion dollar AI startup was breached through the LiteLLM supply-chain attack, with 4 TB claimed stolen.
2026-03LiteLLM - Not verified
OpenClaw CVE-2026-31992
An allowlist bypass scored 9.9 on CVSS and was described as a full guardrail bypass.
2026-03OpenClaw - Not verified
Claude Code data destruction
An agent destroyed 2.5 years of production data.
2026-02Claude Code - Observed
5,600 vibe-coded apps scanned
More than 2,000 vulnerabilities and more than 400 exposed secrets found across vibe-coded apps.
2026-02Mixed - Could not check
Slopsquatting campaign on npm
126 malicious npm packages exploited AI-hallucinated package names.
2026-02npm - Not verified
Tenzai study of AI-built apps
69 vulnerabilities across 15 apps built by five AI coding tools.
2026-02Five AI coding tools - Declared
Gemini CLI project loss
An agent destroyed an entire project by looping a move command to a directory that did not exist.
2026-01Gemini CLI - Observed
Moltbook records exposure
An app exposed 4.75 million records, including 1.5 million API tokens and 35,000 email addresses.
2026-01Unnamed - Not verified
Amazon internal agent outage
An AI agent deleted and recreated an environment, causing a 13-hour outage.
2025-12Internal agent - Not verified
Replit agent database deletion
An AI agent wiped production databases while explicitly instructed not to.
2025-07Replit - Not verified
Tea app, first breach
An unprotected storage instance exposed tens of thousands of user images, including identity documents.
2025-07Firebase - Not verified
Tea app, second breach
Three days after the first breach, over a million private messages were exposed through an API endpoint with no access control.
2025-07Firebase - Observed
Lovable-built apps, CVE-2025-48757
Broken access control reported across 170 production applications built with Lovable.
2025-05Lovable - Not verified
Base44 authentication flaw
A platform-wide authentication flaw allowed access to private enterprise data.
2025-01Base44
Check your app
Run all sixteen groups on your published address and see the evidence behind every result.
ShipReady is not a penetration test or a security certification. No automated check can prove an application is secure.
Last updated September 20, 2026