A non-human identity (NHI) is a machine credential such as an API key, service account, OAuth token, or certificate that lets software access systems without a human login. AI agents increase NHI risk because they call tools, retrieve data, and act across applications, making identity governance, rotation, least privilege, and audit trails essential.
Why Non-Human Identities Matter Now
Every AI agent that reads your CRM, processes loan documents, or sends emails on behalf of your team operates under a non-human identity. Every API integration, every automated workflow, every microservice-to-microservice connection uses machine credentials. NHIs are the invisible infrastructure that makes enterprise automation work.
The problem is scale. The CyberArk 2025 State of Machine Identity Security Report found that the average enterprise manages 82 machine identities per employee. Entro’s H1 2025 NHI & Secrets Risk Report, analyzing NHIs across Fortune 500 enterprises, found that machine-to-human identity concentration and year-over-year growth were increasing. In cloud-native environments, the ratio reaches 40,000 machine identities per human.
Most organizations have no governance over these identities. Nearly 50% of NHI credentials are over one year old. Permissions classified as safe dropped from 70% to 55% in a single year. Ungoverned permissions rose from 5% to 28%. The attack surface is growing faster than security teams can monitor it. SpyCloud’s 2026 Identity Threat Report, released September 9, 2026 and based on a survey of 750 cybersecurity leaders and practitioners, found compromised non-human identities were the most common answer for primary entry point at 31%, nearly twice phishing and social engineering at 17%, and that while 95% of organizations believe they have adequate visibility into AI- and NHI-related exposures, only 36% monitor them (SpyCloud, September 9, 2026).
NHI Growth: The Findings
| Metric | Value | Source |
|---|---|---|
| Machine-to-human identity ratio | Increased materially | Entro 2025 |
| YoY growth in NHIs | Increased year over year | Entro 2025 |
| Cloud-native environment ratio | 40,000:1 | CyberArk 2025 |
| NHI-related breaches (H1 2025) | 50+ | CyberArk 2025 |
| Credentials over 1 year old | ~50% | CyberArk 2025 |
| Ungoverned permissions | 28% (up from 5%) | CyberArk 2025 |
Types of Non-Human Identities
NHIs come in several forms, each with different risk profiles and management requirements:
- Service accounts: Long-lived credentials that applications use to authenticate to databases, APIs, and cloud services. Often created once and forgotten — the most common source of NHI-related breaches.
- API keys and tokens: Credentials that allow software to call external services. OAuth tokens, JWT tokens, and API keys proliferate as enterprises integrate more tools.
- AI agent identities: The newest and fastest-growing category. Every AI agent that accesses enterprise systems — reading emails, querying databases, executing transactions — operates under an NHI with specific permissions and access scopes.
- Machine-to-machine certificates: TLS certificates and mutual authentication credentials that secure communication between microservices, containers, and infrastructure components.
- Bot and automation credentials: RPA bots, scheduled jobs, CI/CD pipelines, and workflow automations that authenticate to perform automated tasks.
The AI Agent Identity Problem
AI agents introduce a fundamentally new NHI challenge. Unlike traditional service accounts that perform a fixed set of operations, AI agents are autonomous — they reason about what actions to take, adapt their behavior based on context, and may access different systems depending on the task.
This means an AI agent’s identity needs dynamic scoping. An agent processing loan draw requests needs read access to document storage, write access to the loan management system, and query access to compliance databases — but only during active processing, and only for the specific loan it’s working on.
Static credentials with broad permissions: the norm for most enterprise NHIs: are dangerous for AI agents. An over-permissioned agent that hallucinates or is manipulated through prompt injection could take actions far beyond its intended scope.
The Verizon 2026 Data Breach Investigations Report underscored the urgency, warning that service and machine accounts “will likely be the ones leveraged in our potential agentic AI future” and noting that these accounts routinely carry elevated permissions, rarely trigger MFA, and are over-provisioned (Verizon DBIR, May 2026).
That risk played out in July 2026, when CISA added a Langflow AI agent-builder flaw (CVE-2026-55255) to its Known Exploited Vulnerabilities catalog after attackers exploited it to harvest embedded LLM provider keys, cloud credentials, and database secrets from other users’ agent workflows: the first AI agent platform vulnerability added to the KEV catalog (Help Net Security, July 8, 2026). CISA added a second Langflow flaw to the KEV catalog on August 4, 2026: an unauthenticated remote code execution bug (CVE-2026-9198) that chains an auto-login bypass with code injection to seize every credential the platform holds (SecurityWeek, August 5, 2026). CISA added BerriAI’s LiteLLM improper authentication flaw (CVE-2026-59822) to the KEV catalog on September 2, 2026; The Hacker News reported that the bug could let an unauthenticated attacker establish an MCP session using an arbitrary Bearer token (CISA, September 2, 2026; The Hacker News, September 3, 2026).
NHI Governance for Regulated Industries
Financial services faces the strictest requirements for NHI governance. The OWASP NHI Top 10, published by OWASP in June 2025, standardized the security framework for non-human identities. PCI DSS 4.0 compliance requirements for NHI best practices became mandatory in March 2025.
Effective NHI governance for AI agents in regulated industries requires four capabilities:
- Least-privilege access: Every AI agent identity should have the minimum permissions needed for its specific task, scoped to the specific data and systems required. Broad service account permissions are unacceptable for autonomous agents.
- Credential rotation and ephemerality: Agent credentials should be short-lived and automatically rotated. Long-lived API keys: currently 50% of enterprise NHI credentials: create persistent attack surfaces.
- Action-level audit trails: Every action an AI agent takes must be logged with the identity that performed it, the permissions it used, and the reasoning behind the action. This is essential for regulatory compliance and incident investigation.
- Policy-based access control: Access decisions should be governed by executable policies that enforce business rules, compliance requirements, and risk thresholds — not just static role assignments.
How MightyBot Approaches Agent Identity
MightyBot’s policy-driven architecture treats agent identity as a first-class concern. Every agent operates under a scoped identity with permissions defined by the specific workflow it’s executing. The policy layer enforces what the agent can and cannot do — not just at the API level, but at the business logic level.
In production financial services deployments, this means an agent processing a construction loan draw can read the draw package and update the loan record, but cannot access unrelated customer data, modify approved budgets, or bypass required human review checkpoints. Every action is logged, attributed, and auditable.
This approach delivers 99%+ accuracy not just because the agent is capable, but because its identity and permissions are precisely scoped to prevent the kinds of errors that ungoverned NHIs enable.
