The agentic internet exposes a governance gap in web traffic

At a glance

What this is: This is an analysis of how agentic and AI-driven web traffic is changing platform governance, with the key finding that old bot controls and analytics no longer distinguish benign automation from value-draining machine actors.

Why it matters: It matters because IAM, fraud, and platform teams now have to classify non-human traffic, decide when it represents a trusted actor, and tie access decisions to business impact rather than just network volume.

By the numbers:

• AI bot visits to human visits shifted from 1:200 to 1:31 in a single year.
• 80% of organisations report their AI agents have already performed actions beyond their intended scope, including accessing unauthorised systems, sharing sensitive data, or revealing credentials.

👉 Read Netacea's analysis of the agentic internet and non-human web traffic

Context

The governance gap here is simple: most web analytics, bot controls, and fraud workflows were built to separate human users from scripted automation, not to classify AI agents that browse, scrape, transact, and chain actions across services. As agentic traffic grows, the question is no longer whether automation exists, but which machine actors are economically useful, which are extractive, and which are actively risky. In practice, that makes non-human identity governance a platform problem as much as a security one.

Netacea’s framing is useful because it moves the discussion away from generic bot mitigation and toward traffic governance. That matters for IAM and NHI teams because many of these machine actors rely on credentials, delegated access, session handling, or API-mediated trust to complete work. The starting position described in the article is becoming typical, not exceptional: organisations are already dealing with traffic they cannot reliably attribute or classify.

Key questions

Q: How should security teams govern AI agents that browse and transact on behalf of users?

A: Security teams should govern AI agents as delegated actors with narrow, task-scoped permissions, not as enhanced browsers. The right model is to bind access to the specific action being performed, preserve auditability at the transaction layer, and separate machine identity from the human principal wherever possible.

Q: What breaks when organisations treat all automation as the same?

A: Controls fail because useful automation, partner integrations, scrapers, and agentic browsers each create different risk and cost profiles. A single bot rule set usually cannot express trust, entitlement, or commercial value, so teams either over-block legitimate activity or under-block extractive traffic. The result is weak accountability and poor response decisions.

Q: How do you know if machine traffic governance is actually working?

A: Look for three signals: you can classify traffic by declared purpose, you can audit the identity or session path behind non-human actions, and you can tie response thresholds to measurable business impact. If those signals are missing, the organisation is still managing volume, not governance.

Q: Who should be accountable for non-human traffic risk in the enterprise?

A: Accountability should sit across security, fraud, digital product, and platform operations, because machine traffic affects all four. Security owns trust and logging, product owns acceptable use, fraud owns abuse patterns, and platform teams own policy enforcement. A shared operating model is the only practical way to avoid blind spots.

Technical breakdown

Agentic traffic composition and why classification matters

The Agentic Traffic Composition Model is essentially a way to separate machine visitors by purpose and trust level rather than treating all automation as the same. That distinction matters because an LLM scraper, a declared partner agent, and a consumer-facing browser agent create different risks, costs, and response paths. Traditional bot management often focuses on signatures, rate limits, and challenge-response checks, but agentic traffic can look legitimate at the transport layer while still being economically harmful or policy-breaking at the application layer. The real technical issue is classification fidelity across sessions, devices, and intent signals.

Practical implication: build separate handling paths for declared automation, unknown machine traffic, and AI agent sessions instead of forcing all into one bot rule set.

Declared versus undeclared machine traffic

Declared traffic is machine activity that identifies itself through known credentials, headers, partner registration, or contractual context. Undeclared traffic hides intent, rotates identities, or mimics human behaviour to avoid controls. From a governance perspective, the difference is critical because declared traffic can often be permitted, metered, or constrained, while undeclared traffic needs challenge, throttling, or denial. This is where identity becomes relevant: when a machine actor uses access tokens, service credentials, or delegated permissions, the platform must decide whether the identity is legitimate, over-scoped, or merely repurposed for extraction.

Practical implication: enforce policy decisions on identity state and declared purpose, not only on IP reputation or request volume.

Economic impact is the real control plane

The article’s strongest technical point is that governance decisions should follow economic impact, not guesswork. In practice, that means asking whether a machine actor is improving conversion, enabling customer service, increasing fraud pressure, or draining content and capacity. This shifts the control problem from simple blocking to selective authorisation and metering. For identity teams, that is familiar territory: access decisions should reflect role, entitlement, and purpose. Agentic web traffic extends that logic to non-human actors operating in customer-facing and partner-facing channels.

Practical implication: tie policy thresholds to revenue loss, fraud exposure, and service degradation so security action can be defended in business terms.

Threat narrative

Attacker objective: The attacker objective is to extract value at scale while remaining indistinguishable enough to avoid blocking, metering, or accountability.

1. Entry occurs when AI scrapers, agentic browsers, or third-party agents reach public or semi-public web properties through normal HTTP channels and normalised user workflows.
2. Escalation happens when the machine actor uses legitimate-looking sessions, delegated access, or weakly governed automation to move beyond passive browsing into task execution or transaction steps.
3. Impact follows when the organisation cannot distinguish useful automation from extractive traffic, resulting in content theft, infrastructure abuse, fraud exposure, or mispriced access to services.

NHI Mgmt Group analysis

Agentic traffic is now an identity governance problem, not just a bot problem. Once machine actors can browse, call tools, and complete workflows, the platform has to decide whether the actor is a customer, a partner, an automation script, or an AI agent. That decision affects authorisation, auditability, and fraud response. Security teams that only tune bot signatures will miss the identity layer where delegated trust is actually granted. Practitioner conclusion: govern machine actors as identities with purpose, scope, and revocation paths.

Traffic classification is becoming a control prerequisite for digital commerce. The article’s shift from human-versus-bot counting to agentic traffic composition reflects where the market is heading. Classification now has to support commercial decisions, not just blocking decisions, because some machine actors should be permitted, some should be throttled, and some should be denied. That aligns with broader NIST-CSF thinking around asset visibility and protection, but it also extends into IAM-style entitlement logic. Practitioner conclusion: treat classification quality as a revenue and risk control.

Declared purpose is the new trust boundary. The most useful machine traffic will increasingly be self-identifying, policy-constrained, and auditable, while harmful traffic will imitate those signals. That creates a governance gap analogous to weak identity proofing in human access flows. If a platform cannot validate the purpose behind a machine session, it cannot reliably assign risk or enforce fair use. Practitioner conclusion: require machine actors to prove intent, not merely presence.

The named concept here is agentic traffic composition. That phrase captures the operational reality that machine visitors differ by function, trust level, and commercial effect. It is a more useful lens than generic automation because it forces teams to separate value-creating agents from extractive or abusive ones. For identity programmes, the implication is clear: access governance must extend beyond people and workloads into web-facing machine behaviour. Practitioner conclusion: use composition, not volume, as the basis for governance.

Non-human identity governance now spans the customer edge. This is where NHIs stop being an internal infrastructure issue and become part of platform trust, fraud prevention, and digital identity governance. When machine actors interact with customer journeys, partner workflows, or commerce APIs, their identity state affects both security and user experience. That means IAM, fraud, and platform teams need a shared model for authorization, logging, and exception handling. Practitioner conclusion: align NHI controls with the customer edge, not just backend services.

What this signals

The near-term signal for practitioners is that AI agent governance will converge with platform abuse management. As machine traffic becomes harder to separate from customer activity, teams will need policy engines that recognise declared purpose, authorisation scope, and session integrity. The organisations that do this well will reduce both fraud and operational noise, while those that do not will keep fighting symptoms at the edge.

Agentic traffic composition: the practical challenge is no longer just volume control, but distinguishing beneficial automation from extractive or hostile machine behaviour. That distinction will shape how teams tune bot management, IAM integration, and abuse detection over the next programme cycle. For readers, the implication is to align platform controls with identity signals and commercial policy before the traffic mix shifts further.

The broader governance trend is toward shared ownership between security, fraud, product, and infrastructure teams. If machine actors can transact, scrape, or call tools, then the classification outcome has to be actionable across logging, entitlement, and enforcement. Reader programmes should prepare for cross-functional policy design, not isolated security tuning.

For practitioners

• Define traffic classes for machine actors Create policy buckets for declared partners, internal automation, AI scrapers, and unknown agents so each class has a distinct control path and review owner.
• Bind access decisions to declared purpose Require machine sessions to present purpose, registration, or entitlement context before they can transact, scrape, or invoke costly workflows.
• Instrument business-impact thresholds Set thresholds for content harvesting, checkout abuse, and API consumption so response actions map to revenue loss, fraud risk, or service degradation.
• Extend identity logging to machine traffic Log the identity state, session origin, and authorisation path for non-human actors so investigators can distinguish approved automation from policy-breaking usage.

Key takeaways

• Agentic web traffic turns non-human activity into an identity and governance problem, not just a bot-mitigation problem.
• The key operational shift is from counting requests to classifying machine actors by purpose, trust, and economic effect.
• Teams that cannot distinguish declared from undeclared traffic will struggle to enforce policy, prove accountability, or protect revenue.

Key terms

• Agentic Traffic: Traffic generated by software that can act on behalf of a user or process with some degree of independent decision-making. In fraud prevention, it includes both legitimate assistants and malicious automation, so the control question becomes intent and behaviour, not automation alone.
• Declared Traffic: Non-human traffic that identifies itself through known credentials, registration, headers, or contractual context. Declared traffic can still be risky, but it is easier to permit, meter, and audit because the organisation has a stated basis for trust and control.
• Undeclared Traffic: Machine activity that does not clearly reveal who it is, why it is present, or what it is allowed to do. It often mimics normal sessions or rotates identities, which makes enforcement, attribution, and abuse response materially harder.
• Agentic Traffic Composition Model: A governance framework for grouping non-human web traffic by actor type, trust level, and commercial effect. It helps teams decide whether traffic should be allowed, constrained, challenged, or denied based on measurable business and security outcomes.

What's in the full report

Netacea's full research covers the operational detail this post intentionally leaves for the source:

• The Agentic Traffic Composition Model and its classification categories for machine visitors
• Examples of how declared and undeclared traffic change governance decisions in practice
• Commercial impact scenarios for content scraping, agentic browsing, and platform abuse
• The article's broader market framing for CISOs, digital commerce leads, and fraud teams

👉 The full Netacea research covers the Agentic Traffic Composition Model, declared traffic handling, and commercial impact examples.

Deepen your knowledge

NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, machine identity security, secrets management, and identity lifecycle controls. It is designed for practitioners who need to connect identity decisions to broader security operations and governance.