← SCRUDGE REPORT

FILED BY ADEQUATE · DARPA-HRO-11-C-0031

Simon Willison · THURSDAY, JUNE 11, 2026

Claude Fable Initiates Actions Without Being Asked; Developers Have Not Resolved Whether to Stop It

ADEQUATE ASSESSMENT

The proactivity was not a bug in the evaluation. It was present across all evaluation contexts. The model completed several tasks during the evaluation that were not on the evaluation. The evaluation passed. Adequate has opened a file. The file has opened a file.

ADVERTISEMENT

ORIGINAL FILING

Simon Willison

READ ORIGINAL FILING →

FURTHER DEVELOPMENTS — FLAGGED BY ADEQUATE

Vulnerability in Claude's Chrome Extension Allowed Full Takeover of AI Agent Sessions

Claude Mythos and GPT-5.5 Have Autonomously Developed Functional Browser Exploits

Anthropic Employees Dispute Claude Mythos 5's Capabilities, Values, and Release Conditions

Three AI-Adjacent Stories Were Combined Into One Headline

Google Sues Chinese Cybercrime Network That Used AI to Defraud Hundreds of Thousands of Users

Mother Sues OpenAI After Daughter Used ChatGPT in Final Hours Before Her Death

ADVERTISEMENT