Anthropic's Code Review leaves preview: AI checks what AI wrote

My take: Claude Code wrote every line of this site, and AI reviewers caught what I missed. I no longer merge AI code without that second read.
What to know
- Code Review left research preview on October 15, 2026, for Claude Team and Enterprise plans, on GitHub only. Several agents read each pull request (a proposed change, waiting for review) in parallel. A verification step drops false alarms, then the findings land as comments ranked by severity. It never approves or blocks a merge.
- At launch, on March 9, 2026, Anthropic said 54% of its own pull requests got substantive review comments, up from 16%, and that fewer than 1% of findings were marked incorrect. A review then averaged $15 to $25 and about 20 minutes; most now take 6 to 10 minutes.
- I built this site with Claude Code on October 25, 2026, and had separate Claude agents review each step: a homemade version of the same idea. They caught an open redirect (a link on my domain that could send you to any site an attacker chose) and two ways to see a newsletter's header image before its send time.
Spotted in TechCrunch, by Rebecca Bellan.
What changes
If your team ships code an AI wrote, an AI reviewer is now a setting, not a project: turn it on where a bug would cost you the most. Keep a person on the merge button, though. A review reads the code without running it, and Anthropic's documentation warns that a review with no findings does not mean the code is secure. On any plan, the /code-review command runs a local version inside Claude Code.
Source: Anthropic, “Code Review is now generally available”, October 15, 2026
Drafted with Claude, Anthropic's AI, from my take. Reviewed and owned by me.