Anthropic makes changes to stop AI agents running amok again

Following three recent security incidents involving Claude, Anthropic is moving to more isolated environments, continuous monitoring, explicit instructions, and proactive human intervention.

Published: 2026-09-01