Skip to content
Hot eventDeveloping

Analysis finds critical security flaws in AI-built apps rise with code size

1 reports1 reporting sourcesUpdated 2 h ago

Start here

AI digest

An analysis of 3,911 public repositories built with nine AI app builders — Bolt, Lovable, Cursor, v0, Replit, Windsurf, Base44, Tempo and Firebase Studio — found that the share of apps with at least one critical security finding rises steadily with code size, from 2.9% for apps under 1,000 lines to 59.2% for those past 40,000 lines. The authors, Path to Ship, also corrected their own headline attribution figure from 92.8% to 78.3% after finding the original calculation circular. The most common issues are missing Supabase row-level security, secrets exposed on the client side, and API routes without authorization checks. The authors note that 66.3% of the corpus has no stars, forks or open issues, and that the study covers only public repositories. The authors say apps of a few hundred lines with no users need not worry, while apps past 10,000 lines with users face a base rate of about 30% that climbs with each feature. They say absolute rates should be read as upper bounds and the trend across code sizes as the main finding.

Generated from reports · updated 2 h ago

Report timeline

Follow the reports to see the event from different sides.

November 20
  1. Hacker News · AI
    Data from 3,911 AI-built repos: critical flaws rise with app size

    An analysis of 3,911 public repositories built with nine AI builders — Bolt, Lovable, Cursor, v0, Replit, Windsurf, Base44, Tempo and Firebase Studio — found the share of apps with at least one critical finding climbs monotonically with code size, from 2.9% under 1,000 lines to 59.2% past 40,000. The authors also corrected their own headline attribution figure from 92.8% to 78.3% after finding it circular, and note that 66.3% of the corpus has no stars, forks or open issues. The three most common issues are missing Supabase row-level security, secrets exposed on the client side, and API routes without authorization checks.

Heat for this event

Not enough continuous observations yet to draw a trend.