Security Anti-Patterns Caused by AI Coding Tools

This report reviews recent incidents where AI-assisted development platforms and integrations caused significant security failures: a Vercel compromise linked to a Lumma infostealer infection and a Lovable regression that exposed public-project source code and AI chat histories due to a Broken Object Level Authorization (BOLA). It then analyzes recurring, platform-level anti-patterns (credential sprawl, permissive RLS, public buckets, webhook handlers without signature verification, hardcoded fallbacks, hallucinated/typosquatted dependencies) that systematically select for insecure defaults and offers practical hardening and incident-response recommendations.