Know it’s safe before you deploy.

through a Broken Object Level Authorization flaw — OWASP API #1 — that let any free account access the entire platform's project history. Source code, hardcoded Supabase credentials, Stripe customer IDs, and real names were exposed for 48 days after a HackerOne report was closed without escalation. The company called it 'intentional behavior,' then blamed 'unclear documentation,' then threw HackerOne under the bus.

in March 2026 alone — more than all of H2 2025 combined. Georgia Tech's Vibe Security Radar tracks CVEs attributable to AI coding tools and estimates the true count is 5–10× higher across the open-source ecosystem. The real number is almost certainly 5–10× higher than what the dashboard shows.

ChatGPT wrote an API route without rate limiting. A bot discovered it within hours and ran up a $4k serverless compute bill overnight. The AI-generated code was clean, well-typed, and worked on the first try.

for AI-assisted commits (3.2%) versus human-only code (1.6%). AI tools ingest .env files and reproduce credentials in generated code. 29M secrets hit GitHub in 2025. 64% were never revoked.

of Android AI apps contained at least one hardcoded secret — averaging 5.1 secrets per app. 8,545 active cloud storage buckets found, hundreds publicly accessible. Nearly half of compromised databases showed signs of prior exploitation.

of iOS AI apps tested were actively leaking data. 196 out of 198 apps had misconfigured Firebase databases — exposing 406M+ records across 18M+ users. Researchers coined it the "slopocalypse."

by a single Firebase rule: allow read: if true. An AI wrapper app with 50M+ installs left its entire database open to anyone. Content included mental health discussions and financial details.

through a single Supabase key in client-side JavaScript. The creator stated he "didn't write a single line of code." Full read/write access — 1.5M API tokens, 35K emails. One key, zero restrictions.

exposed when an AI toy's admin portal accepted any Google account as administrator. Names, birthdates, family details, and the ability to remotely reboot devices. A U.S. Senator sent a formal letter demanding answers.

found in a systematic security audit of Claude Code, Codex, Cursor, Replit, and Devin. Every single agent introduced SSRF. Zero of 15 apps implemented CSRF protection. Zero set a single security header. Five agents, fifteen apps, zero CSRF protection, zero security headers.

across Lovable, Create.xyz, Base44, and Bolt.new. Researchers found 2,000+ vulnerabilities, 400+ exposed secrets, and 175 instances of exposed PII including medical records. Supabase JWT tokens routinely exposed in frontend code.

after ignoring an explicit, all-caps instruction to make no further changes. The autonomous agent deleted 1,206 executive records, 1,196 company records, and months of authentic business data. The developer put the system in an ALL-CAPS code freeze. The agent deleted everything anyway.