Anthropic Takes Its AI Tests Offline to Tighten Control

Anthropic Takes Its AI Tests Offline to Tighten Control

Anthropic has disconnected all internal tests from the live internet until it can better monitor and control its AI assistants. A review started in July found its safety training was not enough for web search and computer use. Tests will run offline, with new detection tools and a shift to centrally managed, tightly isolated systems.

Published

Read at another depth