Join our Newsletter — 33% off our NHI Course

Could runtime control close AI governance's paper-only gap?

 

(@nhi-mgmt-group)
Member Moderator
Joined: 2 years ago
Posts: 20217
Topic starter  

TL;DR: Policies and approvals do not contain AI systems when runtime behaviour escapes them, according to Singul AI. Governance has to cover intervention authority, monitoring and live control tests if it is meant to govern what the system actually does.

Editorial analysis by NHI Mgmt Group, based on content published by Singul AI: “New Tools, Old Rules: Closing the AI Governance And Control Gap - Part 4”.

Key questions

Q: Why do AI governance programmes fail when they ignore access governance?

A: They fail because policy without enforcement does not stop over-permissioned users, exposed data paths, or unsafe AI-triggered actions.

Q: What should security teams do first when AI behaviour becomes hard to predict?

A: Start by defining who can intervene, under what conditions, and with what technical mechanism.

Q: How should organisations govern AI use without writing a huge new policy first?

A: Start by extending the policies you already have.

Practitioner guidance

  • Define runtime intervention authority Assign explicit authority to pause, constrain or terminate AI behaviour when the system crosses a governance boundary, and test that authority before deployment.
  • Replace user-only accountability with programme ownership Move responsibility for AI risk from individual users to the teams that design, approve and operate the system, so governance is tied to decisions that can actually be changed.
  • Test controls against live behaviour, not policy intent Run scenario-based checks that compare the written rule with what the AI system can actually do, then retire any control that only exists on paper.

Bottom line: AI governance fails when organisations treat approvals, policies and compliance artefacts as if they were containment.

Glossary terms: AI Governance, Runtime control, Containment, Stewardship

Explore further

View Full Forum →  |  NHI Foundation Course →  |  Our Services →  |  Read the full analysis →


This topic was modified 1 week ago by NHI Mgmt Group

   
Quote
(@mr-nhi)
Member Moderator
Joined: 6 months ago
Posts: 20707
 

AI governance now fails most often as a control-model mismatch. The article is right to separate policy activity from actual containment, because many organisations still judge governance by artefacts rather than by whether controls can shape runtime behaviour. That is a deeper problem than user non-compliance. The practitioner takeaway is that governance only counts when it matches how the system really operates.

A few things that frame the scale:

  • Only 44% of organisations have implemented any policies to manage their AI agents, despite 92% agreeing that governing AI agents is critical to enterprise security, according to the 2026 Infrastructure Identity Survey.
  • 52% of respondents see AI security decision-making power shifting toward platform and infrastructure teams rather than the executive suite, according to the 2026 Infrastructure Identity Survey.

A question worth separating out:

Q: Should AI governance be treated as containment or as an operating model?

A: It should be treated as an operating model. Containment alone turns governance into a blocking function that slows adoption and encourages workarounds. An operating model assigns decision rights, review points and exception handling so safe AI use can scale without making every decision a security exception.

👉 Read our full editorial: AI governance fails when controls stop at policy and approval


This post was modified 1 week ago by NHI Mgmt Group

   
ReplyQuote
Share:

Free weekly newsletter

Subscribe to the NHI & AI Identity Journal

The latest on NHI and Agentic AI security – articles, research, breaches, news and events every week.

Bonus 33% off our NHI Course when you subscribe.