Development

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

development2026-08-08

Claim

Anthropic rewrote the constitution of Fable 5's biology classifiers to cut benign-query fallbacks by 85% while keeping the dual-use doors locked.

Reported metric: -85% false refusals

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

Sources

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

Welcome to August 8, 2026 (theinnermostloop.substack.com)

Knowledge graphopen in browser →

Reported in

Welcome to August 8, 2026

Actor

Anthropic

Evidence for

Supersedes

Breach responders turn to a rival's open weights because their own models refuse

Referenced by

Cite or correct this record

Citation formatCitation text

A classifier constitution is rewritten to cut benign refusals by 85%
Innermost Loop KB — secondary record
Reporting date: 2026-08-08
KB permalink: https://nicholsn.github.io/innermost-loop-kb/developments/2026-08-08-a-constitution-rewritten-to-cut-false-refusals
Claim: Anthropic rewrote the constitution of Fable 5's biology classifiers to cut benign-query fallbacks by 85% while keeping the dual-use doors locked.
Attributed sources:
- Welcome to August 8, 2026 — https://theinnermostloop.substack.com/p/welcome-to-august-8-2026
This citation does not imply independent verification of linked-source claims.

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