BYOA and trillions of agents will strain enterprise governance

At a glance

What this is: ArmorCode argues that BYOA and agent proliferation will push enterprises toward trillions of autonomous and semi-autonomous agents, turning governance into the harder problem than detection.

Why it matters: For IAM, PAM, and NHI teams, the issue is not just more identities but more self-spawning actors that can access systems, chain actions, and outlive the controls designed for human workflows.

By the numbers:

• Enterprise generative AI adoption crossed 70 percent in under two years, showing how quickly agentic adoption can outpace governance design.
• More than 50 billion connected devices globally create a machine ecosystem large enough to amplify agent-driven exposure across enterprise workflows.
• By 2030, the world may run on 5 trillion agents, according to ArmorCode's analysis of the agentic shift.

👉 Read ArmorCode's analysis of BYOA, shadow AI, and the rise of 5 trillion agents

Context

Agentic AI changes the governance problem because software no longer behaves like a fixed application or a static service account. In BYOA environments, employees can create agents that make decisions, call tools, and move across systems with far less friction than traditional IT deployment.

ArmorCode's core claim is that enterprise risk is shifting from isolated tool misuse to persistent, self-spawning actors that can outnumber human users and conventional applications. That matters to identity programmes because agent access, privilege scope, and lifecycle control become first-class governance issues, not edge cases.

The article treats this as a structural shift rather than a near-term experiment, and that is broadly consistent with what many large enterprises will face: heterogeneous estates, shadow adoption, and long-lived legacy systems will coexist with agentic workflows rather than being replaced outright.

Key questions

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. The control set should include inventory, task-bound credentials, audit trails, and revocation paths. If an agent can call tools or touch production systems, it belongs in the same governance model as service accounts and other machine identities.

Q: Why do APIs create identity risk for IAM and PAM programmes?

A: Because APIs rarely operate without credentials, and those credentials often outlive the service they support. If a token or key is over-privileged, compromise of one integration can expose many systems. IAM and PAM teams need to treat API access as delegated privilege, not just application plumbing.

Q: What do organisations get wrong about shadow AI governance?

A: They often try to block unsanctioned tools at the network layer without changing employee behaviour or providing an approved alternative. That pushes use to personal devices and leaves the enterprise blind. Discovery and policy-guided redirection are more useful than simple denial if the goal is control rather than displacement.

Q: How do security teams reduce risk when agent populations grow faster than controls?

A: Prioritise the agents that can touch sensitive data, administrative systems, or financial workflows, then apply stronger approval, monitoring, and revocation controls to those first. Use identity-centric governance metrics such as ownership completeness, scope accuracy, and closure speed to show whether the programme is keeping up.

Technical breakdown

BYOA creates an identity lifecycle problem, not just an automation problem

Bring Your Own Agent means employees can create agents outside central architecture, then carry them across roles and organisations. These agents may be persistent, short-lived, or event-driven, but each one can hold contextual memory, credentials, and delegated permissions. That makes them closer to a non-human identity than a simple script, especially when they can chain actions across APIs and systems. The governance challenge is lifecycle management: who approved the agent, what identity it uses, what it can reach, and how it is retired when business need ends.

Practical implication: treat every employee-created agent as a governed identity with owner, scope, expiry, and revocation criteria.

Shadow AI becomes a control-plane issue when agents self-spawn at scale

The risk is no longer limited to employees pasting sensitive data into unsanctioned tools. Persistent agents can act without approval at every step, store memory, and interact with enterprise data and services in ways that are difficult to observe using traditional application controls. Because the agent surface expands across development, finance, IT, and security workflows, the problem becomes one of control-plane visibility. Security teams need to know where agents exist, what they inherit, and how their privileges are constrained in runtime.

Practical implication: build discovery and authorization controls for agents before proliferation makes inventory and containment unreliable.

Agentic scale demands governance across ownership, context, and closure

ArmorCode's framing reflects a real operational shift: detection will improve faster than remediation. In agentic environments, the hardest questions are not whether a vulnerability exists, but which agent owns the action, which business context justifies the access, and whether the workflow can be closed safely. That aligns with identity governance principles, but at much higher velocity and lower visibility. NHI controls such as short-lived credentials, scoped delegation, and activity logging become essential because every agent interaction is a potential access event.

Practical implication: align agent governance to identity lifecycle controls, not just scanner outputs or application approvals.

NHI Mgmt Group analysis

BYOA is the point where agentic AI becomes an identity governance problem. Once employees can create their own agents, the enterprise is no longer governing a tool that a central team deployed. It is governing an expanding population of delegated actors with unclear ownership, variable scope, and inconsistent retirement. That means the familiar IAM question, who should have access, now extends to what should exist as an acting identity at all. Practitioners should treat BYOA as a governance boundary event, not an innovation milestone.

Shadow AI is evolving from unsanctioned usage into unsanctioned agency. The first wave exposed data because people typed sensitive content into external systems. The next wave exposes systems because agents can make decisions, retain memory, and execute chained actions across APIs. That shift creates a stronger need for NHI-style controls around credentials, token scope, and revocation. For identity teams, the lesson is clear: unmanaged agent sprawl is a governance debt problem, not merely a user training issue.

Five trillion agents is less a prediction than a capacity challenge for security operations. Whether the eventual count lands exactly there is less important than the directional reality that machine actors will vastly outnumber human users. That changes what the security programme must measure: not just vulnerability volume, but ownership clarity, policy coverage, and closure latency. Teams that still optimise for alert quantity will miss the bigger issue. Practitioner conclusion: build governance models that scale with actor count, not with headcount.

Agentic AI will accelerate convergence between IAM, PAM, and NHI governance. When agents can request data, invoke tools, and chain tasks, the boundary between human delegation and machine operation becomes operationally thin. Identity teams should expect pressure to unify lifecycle control, privilege limitation, and runtime monitoring across all actor types. That does not mean collapsing every control into one tool; it means forcing one governance model to describe every actor that can act on behalf of the enterprise. Practitioners should prepare for policy unification, not control fragmentation.

What this signals

Shadow AI is turning into shadow agency. Security programmes that can only count sanctioned tools will miss the operational risk of agents that inherit credentials, retain memory, and invoke APIs independently. The practical response is to extend identity governance into runtime authorisation and closure, using models aligned with NIST AI Risk Management Framework and agent-focused controls such as OWASP Agentic AI Top 10.

Closure speed will become a core control metric for agentic programmes. If an agent can be created in minutes but decommissioned only through manual coordination, the exposure window is structurally too wide. That is where NHI-style governance, especially the lifecycle thinking in the Ultimate Guide to NHIs, becomes operationally relevant to AI security teams.

The next control gap is not discovery alone, but proving that the enterprise can answer three questions consistently: who owns the agent, what can it reach, and how fast can it be stopped. Teams that cannot answer those questions will struggle to defend the rest of their AI governance programme.

For practitioners

• Define a governed agent inventory Create an inventory of all employee-built and system-built agents, including owner, purpose, data access, tool access, and retirement criteria. Make inventory status a prerequisite for production use, not a documentation exercise.
• Bind agent access to short-lived credentials Issue scoped, short-lived credentials or delegated tokens for agent workflows so access expires with the task, session, or approval state. Avoid persistent credentials that let agents continue operating after business need has ended.
• Apply privileged access controls to high-impact agents Classify agents that can approve, transfer, delete, or disclose data as privileged actors and subject them to step-up controls, approval paths, and break-glass review. Tie these controls to the systems they can reach, not just to the person who created them.
• Track agent closure latency as a governance metric Measure how long it takes to revoke an agent, disable its credentials, and stop its workflows after role change, project end, or policy breach. Long closure times are a sign that lifecycle governance has not kept pace with agent proliferation.

Key takeaways

• BYOA turns AI agents into governed actors, not disposable automations, which makes identity lifecycle control central to risk management.
• The article's scale argument is that agent populations will grow faster than visibility, making ownership, scope, and closure the real bottlenecks.
• Practitioners should move from tool-level oversight to identity-centric governance for every agent that can access data or chain actions.

Key terms

• Bring Your Own Agents (BYOA): A pattern where employees build and deploy AI agents outside central security review. These agents often connect directly to enterprise systems, creating identity, access, and accountability issues that look like shadow IT but behave more like shadow AI.
• Shadow AI: AI agents, copilots, or connected tools operating without full visibility or governance from security teams. Shadow AI becomes an identity problem when those systems authenticate with unmanaged tokens, service accounts, or OAuth apps that can reach production resources.
• Agentic AI: Autonomous AI systems capable of planning, deciding, and taking actions — including calling APIs, writing code, and orchestrating other agents — with minimal human oversight. Agentic AI introduces new NHI risks as agents must authenticate to external services.
• Verification latency: The time between a claimed fix and independent confirmation that the exposure is actually gone. Long verification latency creates a false sense of closure and extends the period in which teams may think risk has been removed when it has not.

What's in the full article

ArmorCode's full blog post covers the strategic reasoning and directional forecasting this post intentionally leaves at the source:

• The detailed argument behind the five-trillion-agent projection and the assumptions used to build it
• The article's broader view of how M&A-driven complexity and legacy estates change the agentic threat model
• ArmorCode's explanation of why exposure management, not scanner volume, is the bottleneck in agentic environments
• The vendor's framing of aggregate, analyse, and automate as an operating model for the agentic economy

👉 ArmorCode's full blog post expands on the agentic-era governance model and the operating assumptions behind its forecast.

Deepen your knowledge

The NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, agentic AI identity, and machine identity security. It is designed for practitioners who need a common control language across IAM, PAM, and autonomous systems.