Arkose Labs adds session policies and Agent ID to Agent Trust Manager
Arkose Labs has expanded its Agent Trust Manager with controls for classifying and governing agent sessions across customer-facing workflows. Announced November 5, 2026, the additions are intended to let businesses decide how different types of agents may interact at stages such as signup, login, checkout and money movement.
The product addresses a growing identity challenge: automated agents may not reliably identify themselves, while authentication and fraud controls designed for people may not distinguish an agent acting on a customer’s behalf from other automated activity. Arkose Labs says its State of Agentic AI Trust Q3 report found that personal AI agents identify themselves less than 3% of the time. The company argues that policies relying only on an agent’s self-declaration will therefore miss most agents.
Policies across the customer journey
With the expanded orchestration capability, organizations can map steps in a customer journey and assign an allow, monitor or block policy to each one. Arkose Titan applies those decisions during an interaction, according to the company. The approach gives teams the option to permit an agent in one context while monitoring or denying it in another, rather than applying a single decision to every interaction.
Agent Trust Manager classifies sessions as self-disclosing good, non-disclosing good or malicious. The classifications are intended to help businesses distinguish agents that identify themselves from those that do not, and assess whether an agent is acting for a real customer. Arkose says organizations can use the classification to apply policies to agent populations at each workflow stage.
When a business blocks an agent, the system can display a message in the business’s brand explaining that the workflow is not supported for agents and pointing to a supported path. This provides a way to communicate a policy decision rather than returning only an unexplained denial. The source announcement does not specify which customer journeys or integrations are available to individual buyers, or how classifications are validated.
Agent identity and detection records
Arkose is also introducing Agent ID, a stable identifier designed to follow an agent rather than its IP address. The company says the identifier is intended to remain useful when an agent changes networks, ships a new build or appears across multiple accounts. That capability could help security and identity teams correlate activity that would otherwise be separated by changing network details, although the announcement does not describe the identifier’s technical construction or guarantees.
A separate detection-auditing capability will record when detection signals were updated, how sessions were classified at the time, and what a newer signal would have detected. Arkose says the audit trail can be exported from Command Center. Such records are relevant to teams reviewing policy decisions and changes to detection over time; the announcement does not specify retention periods or export formats.
Command Center also provides access to agent traffic information, including agent mix over time, attack campaigns and countries of origin, in comparison with activity across the Arkose network. The company says teams can combine controls with additional intelligence sources and share them internally through Command Center.
AI assistant access and human confirmation
“Ask Titan” and an AI Assist integration using MCP are designed to provide plain-language answers about agents in a business’s traffic, either in Command Center or through an AI assistant the organization already uses. Arkose says access is governed by existing roles. The announcement does not identify supported assistants or provide further details about the MCP implementation.
Arkose also described a “Proof of human” capability as coming soon. Businesses will be able to mark sensitive steps, such as moving money, as requiring a person to confirm before an agent proceeds. This is presented as an additional control for selected actions, not a blanket restriction on agent use.
For enterprise identity and fraud teams, the expanded product brings together session classification, workflow-level decisions, cross-network identification and an audit trail. Organizations evaluating these controls will need to determine how agent categories align with their own customer policies, which actions warrant human confirmation, and how automated decisions should be reviewed. Arkose Labs CEO Kevin Gosschalk said businesses can no longer rely on agents to announce themselves, citing agents’ ability to change signals and pass authentication steps built for people.
