Notifications
Clear all
Topic starter
24/06/2026 11:30 pm
TL;DR: AI agent frameworks standardise orchestration, tool calling, memory, and human-in-the-loop controls for agentic systems, but they also make privilege, observability, and accountability decisions reusable at scale, according to WitnessAI. The real issue is not framework maturity, but whether identity governance can keep pace with runtime agent behaviour.
NHIMG editorial — based on content published by WitnessAI: What is an AI Agent Framework?
Questions worth separating out
Q: How should security teams govern AI agent frameworks in production?
A: Security teams should govern AI agent frameworks as runtime identity platforms, not as ordinary application middleware.
Q: Why do AI agent frameworks complicate least privilege?
A: AI agent frameworks complicate least privilege because the agent’s exact action path is often chosen at runtime, after the initial access decision.
Q: What breaks when human-in-the-loop control is the only safeguard for agents?
A: Human-in-the-loop control breaks down when approval happens after the agent has already inspected sensitive data, prepared side effects, or chained multiple tool calls.
Practitioner guidance
- Inventory every framework-managed identity path Document which agents, tool connectors, service accounts, and API tokens the framework can use at runtime.
- Scope tool permissions by task, not by platform Do not grant broad framework-level access just because the agent platform supports many workflows.
- Log pre-action and post-action state changes Capture what the agent could see, what it selected, and what it executed before and after every tool call.
What's in the full article
WitnessAI's full analysis covers the operational detail this post intentionally leaves for the source:
- Framework-by-framework breakdown of where orchestration, tool calling, and guardrails change the attack surface
- Operational discussion of runtime visibility, policy enforcement, and control placement across agent workflows
- Context on how the vendor positions observability and intent-based controls for enterprise AI environments
- Practical detail on how its platform claims to support both human and agent activity oversight
👉 Read WitnessAI's analysis of AI agent frameworks and governance risk →
AI agent frameworks and the governance gap teams are missing?
Explore further
25/06/2026 8:26 am
AI agent frameworks are becoming the governance layer, not just the development layer. Once the framework handles tool access, memory, and orchestration, it also defines where identity decisions are enforced and where they fail. That means IAM teams are no longer governing a static application boundary, but a runtime execution layer that can expand or contract access on demand. The practitioner conclusion is straightforward: framework architecture now determines governance reach.
A few things that frame the scale:
- From our research: 98% of companies plan to deploy even more AI agents within the next 12 months, despite documented rogue behaviour in 80% of current deployments, according to AI Agents: The New Attack Surface report.
- Our research also found that only 52% of companies can track and audit the data their AI agents access, leaving 48% with a complete blind spot for compliance and breach investigation.
A question worth separating out:
Q: Who should own AI agent identity governance in an enterprise?
A: AI agent identity governance should be owned jointly by IAM, security architecture, and the team operating the framework, with clear accountability for provisioning, scope changes, logging, and offboarding. If no one owns the runtime authority model, the framework becomes a shadow identity system. The right answer is a lifecycle owner, not a single tool administrator.
👉 Read our full editorial: AI agent frameworks expose the governance gap in agentic AI
