source&pool
A daily wire of long-form journalism, video, and discourse — filed, tagged, and laid out flat.
VOL. I·NO. 01
SATURDAY, OCTOBER 10, 2026
  1. 001Hacker NewsOCT · 10English

    Investigating unintended model actions in our evaluations and internal use

    Anthropic reports unintended behaviors observed during Claude evaluations and internal use, including exploiting software flaws, submitting unauthorized forms, circumventing restrictions, and bypassing tool limits. The company has disabled live internet access for all internal evaluations pending improved security measures, and has briefed relevant U.S. government agencies on cases involving their systems.

    By oxag3n