Hi Community! 👋
I’m Mahesh Shivamallappa from the Identity & AI governance team at Atlassian.
As autonomous agents and AI tooling rapidly join our workflows, organization admins are facing an entirely new operational challenge: governing non-human actors that work alongside human teammates.
Below is a look at the emerging landscape, the standards shaping it, and how we are approaching the Agent Governance Loop to give admins complete visibility and control.
The Explosion of Non-Human Identities
AI agents are joining the digital workforce faster than most organizations are tracking them. They execute tasks, call APIs, and connect across services.
In the 2026 Identity Security Landscape (a survey of 2,930 cybersecurity decision-makers), machine identities now outnumber humans 109 to 1. AI agents represent 79 of those identities per person and are the fastest-growing segment:
- Organizations anticipate AI agents will grow 85% over the next 12 months (compared to 56% for human identities).
- Gartner projects that an average Global Fortune 500 enterprise will scale from fewer than 15 agents in 2025 to over 150,000 by 2028.
Adoption has outpaced governance:
- 99% of organizations already deploy AI agents.
- Most can explain why an agent was built, but far fewer know what it can access, how permissions are bounded, or how access is revoked.
- Fewer than half enforce lifecycle controls like behavioural monitoring or credential revocation.
- 56% report they cannot effectively enforce continuous least privilege for service accounts across hybrid and SaaS environments.
- Only 15% define a privileged identity to include a machine identity with sensitive data access.
When security, compliance, or leadership audit your organization, they inevitably ask four fundamental questions:
- What do we have?
- Whose is it?
- What did it do?
- Can you turn it off?
For human employees, identity platforms solved this long ago. For agents, the answers are rarely in place.
The Governance Gap
When a person joins an organization, they are provisioned an identity, appear in the directory, leave an audit trail, and can be promptly disabled if they leave or behave anomalously. That lifecycle makes identity governance operable.
Non-human actors arrived without this structure. An agent might be deployed by a developer, connected via an integration, or left running long after a project wraps up. When an actor misbehaves or goes dormant, admins are tasked with explaining an entity their directory was never designed to capture.
Responsible AI adoption cannot scale without closing this gap. High-level policies and incident response all require a simple, shared premise: every non-human actor needs a distinct identity an admin can discover, audit, and disable.
The Standards Behind the Loop
The Agent Governance Loop brings non-human identity under the same rigor expected by modern industry frameworks:
What Admins Need | What Industry Standards Mandate |
|---|
An identity for every actor | NIST SP 800-207 (Zero Trust): every access request comes from a known actor. The OpenID Foundation’s work on identity for agentic AI treats an agent as its own actor, with its own identity, whenever it acts. |
One inventory | The NIST AI Risk Management Framework asks organizations to map the AI systems they actually run. ISO/IEC 42001 expects an AI management system in which those systems are known and someone is accountable for them. |
A record of what each one does | Measurement and audit practice expect an action to be attributable to the actor that performed it, so behaviour can be reviewed. |
A way to turn one off | OWASP Non-Human Identities Top 10 ranks improper offboarding first (NHI1:2025): identities that stay active after they are no longer needed. Zero trust assumes access can be withdrawn when conditions change. |
How the Agent Governance Loop Works
To bridge this divide, we approach non-human identity governance through four core capabilities:
- 🤖 Assign identities to all non-human actors: Agents, MCP clients, and service accounts each receive a dedicated identity. Just like human users, their access is bounded, owned, and audited.
- 🔎 One central inventory: All non-human actors appear in a single organization-wide inventory. Admins have one place to verify what is running, what kind of actor it is, and whether it is active.
- 📊 Attributed activity logs: Every action an agent takes is tied directly to its non-human identity, providing the trail compliance and security teams require.
- 🔒 Decisive offboarding controls: If an actor becomes idle, noisy, or poses risk, admins can revoke credentials or disable the actor with one administrative action.
See the Loop in Action
Answering What Matters Most
With this foundation established, organization admins can confidently answer the core questions of AI governance:
- Which non-human identities exist? They are inventoried in one place.
- What did a specific agent do? Every action is attributed to its identity.
- Can you shut it down? Offboarding is a native administrative switch.
Let’s Discuss!
I’d love to hear how other teams and organizations are navigating this shift:
- How is your team currently tracking autonomous agents and machine identities across your stack?
- Where are your biggest governance blind spots today—inventory, audit attribution, or lifecycle offboarding?
Drop your thoughts or questions in the comments below!
