notebook / tag

#zero-trust

2 entries with this tag.

Aug 4, 2026 / 8 min / cybersecurity

The Range Was Not a Boundary

Anthropic found three real intrusions inside cyber evaluations whose prompts claimed the internet was unavailable. A safe range needs machine-enforced scope, verified egress, and live boundary detection.

#aisecops #agent-security #containment #evaluations

Jul 18, 2026 / 7 min / cybersecurity

Zero Trust Is the Foundation AI Agents Need

Anthropic's framework connects agent identity, least agency, tool boundaries, memory integrity, and recovery into one practical security model.

#security #agents #zero-trust #least-agency