EDITORIAL BRIEF / NOVEMBER 2026
Map the decision right
A tool call can change inventory, disclose a record or make a commitment. Identify the authorised actor, the permitted operation and the action’s impact before allowing write access. Give each tool the narrowest permissions necessary and enforce them outside generated text.
Make approval part of execution
Tie approval to a specific operation and version of the record. If the proposed action changes after approval, return it for review. Log actor, evidence, inputs, outcome and escalation. Test revocation and rollback before expanding access.
What the market signals
ServiceNow’s AI Control Tower announcements and AI Passport proposal illustrate the growing focus on visibility and runtime authority across vendor boundaries. These are vendor-reported capabilities. Buyers should independently test whether policies are enforced across each connector and workflow.
Primary sources
- NIST AI Risk Management Framework · January 2023
Voluntary framework for governing, mapping, measuring and managing AI risk.
- OWASP Top 10 for Agentic Applications for 2026 · 9 December 2025
Practitioner taxonomy of risks specific to AI systems that take actions across workflows.
- ServiceNow expands AI Control Tower · 5 May 2026
The company announced cross-platform discovery, observation, governance, security and measurement capabilities.
- ServiceNow announces AI Passport · 15 October 2026
The company described runtime authorization for first- and third-party agents and MCP servers.