Analysis

A Forbes Council essay raises a useful enterprise-security question: how well can an organization understand and stop an agent’s actions? The question reaches beyond detecting an error to the permissions, oversight and accountability built into a deployment.

Separate joint guidance from the Canadian Cyber Centre and partner agencies in Australia, the United States, New Zealand and the United Kingdom recommends caution with agentic AI. It calls for assessing operational risk, maintaining visibility and assurance, and aligning controls with the organization’s existing security model. It advises against broad access to sensitive data or critical systems and favors low-risk, non-sensitive uses.

A useful pilot should have a deliberately limited job, a named owner and an explicit way to stop access. A team can review what the agent may read, which tools it may use and what requires human approval before expanding the task. The guidance provides a practical framework for that assessment; it does not establish that a particular organization has suffered an incident or that a specific product makes a deployment safe.

Source: Reported headline source: forbes.com · 2026-11-06

Source: Primary source: cyber.gc.ca

Why it matters

A bounded pilot makes responsibilities and exposure easier to assess than an open-ended permission grant; it does not remove the need for continuing oversight.

About this article

AI-assisted Silver Owl Journal editorial analysis. Named editorial personas are not human journalists.

Editorial standards & AI production ↗ · About Theo Park

Sources and editorial note

Approved November 7 manual edition. Source attribution and evidence scope are stated in the analysis.

  • Reported headline source: forbes.com (2026-11-06)
  • Primary source: cyber.gc.ca