Notifications
Clear all
Topic starter
16/09/2026 9:21 am
TL;DR: A core agentic security problem is underscored by WitnessAI’s inclusion in Software Analyst Cyber Research’s ARISE map: valid credentials can still drive invalid actions unless identity, intent and data context are evaluated together, with inline intervention before completion. The real governance gap is assumption collapse, because access approval alone cannot govern runtime tool use or external data transfer.
NHIMG editorial — based on content published by WitnessAI: ARISE market map coverage and runtime agent governance analysis
Questions worth separating out
Q: How should security teams govern AI agents that can access enterprise systems?
A: Security teams should govern AI agents as non-human identities with explicit ownership, scoped privileges, and continuous monitoring.
Q: Why do shared credentials create risk in agentic workflows?
A: Shared credentials erase the line between requester and operator, so security teams cannot tell whether a Lambda invocation, database query, or EC2 session came from the right agent and task.
Q: What breaks when AI agent governance is only detection-based?
A: The control arrives after the action has already happened, which means sensitive transfer, misuse, or policy violation may already be complete.
Practitioner guidance
- Define runtime decision boundaries Map the exact point where an AI agent may proceed, pause, redact, or stop based on identity, intent, and data classification before an external tool call completes.
- Instrument tool calls inline Place controls in the traffic path for prompts, MCP connections, and downstream actions so policy can affect the event before the external transfer occurs.
- Separate visibility from enforcement Test whether the platform can actually block or restrict an action rather than only detect it after the fact, and document which workflows remain alert-only.
What's in the full article
WitnessAI's full article covers the operational detail this post intentionally leaves for the source:
- How the ARISE control-depth model distinguishes visibility from inline enforcement across agent workflows.
- The four ARISE placement areas and what each implies for policy decision points in runtime.
- How WitnessAI evaluates prompts, MCP connections, and tool calls in the traffic path.
- The buyer questions SACR used to test whether a platform can actually decide, intervene, and generate evidence.
👉 Read WitnessAI's analysis of ARISE runtime governance for AI agents →
AI agent tool governance: are your controls deciding on action, not just access?
Explore further
16/09/2026 9:21 am
Agentic security is an action problem, not an access problem. The article shows that a valid credential can still drive an invalid action, which is the central governance break for AI agents. Traditional IAM answers who may enter a system, but runtime agent governance has to answer whether a specific tool call should proceed with a particular data set and purpose. The practitioner conclusion is that permissioning alone no longer defines control for autonomous work.
A few things that frame the scale:
- 85% of organisations lack full visibility into third-party vendors connected via OAuth apps, according to The State of Non-Human Identity Security.
- 38% have no or low visibility into those OAuth-connected vendors, which shows how quickly delegated access can outpace governance.
A question worth separating out:
Q: How do human accountability and AI agent execution stay connected?
A: They stay connected when the employee, the task, and the agent activity are recorded as one governance chain. That allows investigators to see who delegated the work, why it was requested, what data was touched, and which tool action was attempted. Without that linkage, accountability becomes fragmented.
👉 Read our full editorial: Agentic runtime governance needs identity, intent and data together
