Development

An agent's auto mode blocks harmful actions most paid humans allowed

development2026-08-10

Claim

Anthropic is making auto mode the Claude Code default after evaluations showed it blocked 89% of harmful actions that only 13.6% of paid humans refused.

Reported metric: 89% vs 13.6%

Reporting date: 2026-08-10 (not necessarily the event date)

Sources

Attribution and reporting linked by this record; links do not imply independent verification.

Welcome to August 10, 2026 (theinnermostloop.substack.com)

The model refuses what most of the humans hired to check it did not.

Knowledge graphopen in browser →

Reported in

Welcome to August 10, 2026

Actor

Anthropic

Evidence for

Supersedes

A classifier constitution is rewritten to cut benign refusals by 85%

Referenced by

Welcome to August 10, 2026

Cite or correct this record

Citation formatCitation text

An agent's auto mode blocks harmful actions most paid humans allowed
Innermost Loop KB — secondary record
Reporting date: 2026-08-10
KB permalink: https://nicholsn.github.io/innermost-loop-kb/developments/2026-08-10-auto-mode-blocks-what-humans-would-not
Claim: Anthropic is making auto mode the Claude Code default after evaluations showed it blocked 89% of harmful actions that only 13.6% of paid humans refused.
Attributed sources:
- Welcome to August 10, 2026 — https://theinnermostloop.substack.com/p/welcome-to-august-10-2026
This citation does not imply independent verification of linked-source claims.

Suggest a correction — opens a GitHub draft for you to review and submit.