Customer AI agent

Expanded Definition

A customer AI agent is not just a conversational assistant. It is a delegated software actor that can browse, compare, fill forms, hand off context, and sometimes complete transactions on a person’s behalf. In NHI terms, that means it behaves like a non-human identity with a human sponsor, a bounded purpose, and policy constraints tied to the customer relationship.

Definitions vary across vendors and product teams, because some organisations treat these agents as a chatbot feature while others treat them as autonomous delegates with execution authority. NHI Management Group recommends the latter view when the agent can access accounts, authenticate, or trigger business actions. That framing aligns with the OWASP Agentic AI Top 10 and the governance model in the NIST AI Risk Management Framework.

A customer AI agent differs from a standard bot because it is expected to act with delegated intent, not merely respond to prompts. It also differs from internal enterprise agents because the trust boundary includes customer consent, customer data, and downstream commercial consequences. The most common misapplication is treating a customer AI agent as a simple UX layer, which occurs when teams ignore its authentication scope, tool access, and audit requirements.

Examples and Use Cases

Implementing customer AI agents rigorously often introduces consent, traceability, and fraud-screening overhead, requiring organisations to weigh convenience against stronger identity controls and more detailed logging.

• A retail agent searches product catalogues, compares shipping terms, and purchases an item under a customer-approved spending limit.
• A travel agent signs into a booking portal, uses stored preferences, and books flights or hotels while preserving an auditable action trail.
• A banking assistant drafts a payment initiation or card dispute, but stops short of release until the customer confirms the transaction.
• A healthcare scheduling agent exchanges limited booking data with a provider system, but only after explicit customer consent and scope checks.
• Research on agent abuse patterns in AI Agents: The New Attack Surface report and OWASP NHI Top 10 shows why these use cases need policy boundaries, not just conversational guardrails.
• Designers commonly reference CSA MAESTRO agentic AI threat modeling framework when mapping tool access, escalation, and delegation flows.

Why It Matters in NHI Security

Customer AI agents expand the attack surface because they can inherit customer authority, move across systems, and expose secrets, tokens, or personal data if delegated access is too broad. NHIMG research shows that only 52% of companies can track and audit the data their AI agents access, while 80% report agents performing actions beyond intended scope. That gap turns customer-facing automation into a governance problem as much as a technical one.

This is also where identity and secrets management intersect. If an agent can retrieve or relay credentials, the organisation must assume it can leak or misuse them under prompt manipulation, account takeover, or over-permissioned tool access. The secrets risk profile described in The State of Secrets in AppSec becomes relevant when agents touch APIs, session tokens, or delegated authentication flows.

NIST guidance on identity assurance and risk handling through the NIST AI Risk Management Framework helps organisations frame this as a lifecycle control problem, not a one-time product decision. Organisations typically encounter the consequences only after a customer agent has overreached, misrouted data, or completed an unintended transaction, at which point customer AI agent governance becomes operationally unavoidable to address.

Related resources from NHI Mgmt Group

• AI Agent Authentication
• Who should be accountable when an AI marketing agent changes customer data incorrectly?
• Who should approve AI agent access before customer-facing deployment?
• Shadow AI