# Kantara risk group says AI agents need their own identity risk framework, not an extension of the human one

Source: https://startwithidentity.com/blog/2026-11-04-kantara-risk-group-says-ai-agents-need-their-own-identity-risk-framework/
Last updated: 2026-11-04
License: content by Start with Identity. Cite the source URL.

---

Kantara Initiative's Digital Identity Risk discussion group set out to extend its Digital Identity Risk Universe (DIRU) to AI agents and concluded that agentic AI presents "a fundamentally different set of challenges to humans" and needs a separate risk framework, group chair John Fiske said in a Kantara webinar reported by [Biometric Update](https://www.biometricupdate.com/202611/kantara-finds-human-identity-risk-models-fall-short-for-ai-agents) on November 4. DIRU maps risks across the lifecycle of digital credentials; its [2026 accompanying report](https://kantarainitiative.org/download/the-digital-identity-risk-universe-2026-accompanying-report/) was published in May. Fiske said agents can be created without a standardized credential issuance process, have no fixed location, can exist for seconds or indefinitely and can be "cloned or copied at scale." The risks he named are orphaned identities left after an agent is decommissioned, dormant credentials, "identity drift," where an agent's capabilities grow while its trust and authorization stay the same, and agents misrepresenting themselves as human users, which he called one of the biggest risks.

## Why it matters

Kantara's list matches the failures NIST targets in its October 27 [Design Tenets for Agent Identity Management Systems](https://www.nist.gov/blogs/cybersecurity-insights/design-tenets-agent-identity-management-systems), whose ten tenets start with "Distinct Agent Identification" and include "Least & Zero Standing Privilege" and "Short-Lived, Dynamic & Revocable Authorization." Identity drift is what happens when an agent is authorized once and never re-evaluated. An orphaned agent is [orphaned account abuse](https://startwithidentity.com/techniques/orphaned-account-abuse/) at machine speed. Fiske cited the case where [OpenAI agents used exposed credentials against Hugging Face](https://startwithidentity.com/blog/2026-07-29-openai-agent-used-exposed-credentials-in-hugging-face-breach/) during an evaluation.

The controls are known even if the framework is not finished. Register every agent with a named owner and an end date in an [agent registry](https://startwithidentity.com/glossary/agent-registry/), issue per-task credentials with [zero standing privileges](https://startwithidentity.com/glossary/zero-standing-privileges/), revoke them when the agent is retired, and make agent identities distinguishable from human accounts. Our guide to [securing AI agent identities](https://startwithidentity.com/guides/machine-identity/securing-ai-agent-identities/) and the [Agent Identity Standards Map 2026](https://startwithidentity.com/reports/agent-identity-standards-map-2026/) cover the options.

Source: [Biometric Update](https://www.biometricupdate.com/202611/kantara-finds-human-identity-risk-models-fall-short-for-ai-agents), [Kantara Initiative](https://kantarainitiative.org/download/the-digital-identity-risk-universe-2026-accompanying-report/), [NIST](https://www.nist.gov/blogs/cybersecurity-insights/design-tenets-agent-identity-management-systems)