SailPoint report reveals AI agent security lag behind adoption

SailPoint released its Horizons of Identity Security report, revealing a velocity paradox between AI agent deployment and human-speed identity controls.

According to the report, seventy-nine percent of enterprises are running AI agents in production, while only two percent have deployed identity‑security tools purpose‑built for governing those agents. This represents a roughly forty‑to‑one gap between AI agent adoption and purpose‑built identity security controls.

About sixty percent of organizations remain in the foundational stages of identity security, corresponding to Horizon 1 or Horizon 2. Five years ago, forty-five percent of organizations were at the lowest maturity level (Horizon 1) for human identity security; today that figure has fallen to twenty-three percent. Currently, fifty-four percent of organizations are at Horizon 1 for agent (non-human) identity security.

Less than one percent of organizations have reached Horizon 5 (the highest maturity) in identity security, while roughly sixty percent remain in Horizons 1‑2. Core operational hurdles cited include credential lifecycle management, shadow AI discovery, and real‑time monitoring. SailPoint CTO Chandra Gnanasambandam said that expecting a security admin to review and approve access for autonomous agents making ten thousand tool calls per second is outdated and a fantasy.

Matt Mills, SailPoint President, stated that autonomous agents have moved beyond experimental copilots and now serve as the primary execution layer in modern enterprises. SailPoint introduced Autonomous Agents, specialized AI agents that watch runtime behavior, block threats, and tweak permissions while letting approved agents keep running. The company is introducing Autonomous Identity Security Posture Management (A‑ISPM) that uses Atlas platform data to identify dormant accounts, orphaned access, shadow administrators and other risky paths, enabling remediation such as revoking access, disabling an agent, assigning an owner or automated enforcement.

Just‑in‑Time provisioning grants access only when needed, with conditions such as reauthentication or a business justification, and can be activated via Slack or ServiceNow. The report identifies a digitization trap in which applying scheduled access‑review cycles to short‑lived machine identities creates operational drag and is ineffective. Advanced organizations have moved away from manual, ticket‑based access decisions and now use continuous, contextual, automated policy enforcement that operates at machine speed.

Nearly half of the market (49%) says it balances speed and security equally, but the report characterizes this stance as a false compromise lacking the operational capability to enforce it. AI agents account for about twenty-two percent of all non‑human accounts, yet eighty‑five percent of organizations continue to rely on legacy identity tools not designed for agentic identities (identities of AI agents). Eighty‑seven percent of respondents rate their human IAM capabilities as capable or better, but only forty‑three percent report mature processes for agentic access.

Eighty percent of leaders believe the gap in their current tooling is moderate or smaller, yet only fifteen percent can provision non‑human access in real time. Fifty‑seven percent express confidence in meeting regulatory requirements, while only forty‑three percent feel prepared to produce verifiable evidence for an AI‑related audit. Organizations that have invested in securing non‑human identities report measurable productivity gains (sixty‑two percent) and safer, faster AI deployment (forty‑six percent).

Seventy‑six percent of leaders expect stronger identity governance to improve eligibility for or the terms of cyber insurance. Wendy Wu, SailPoint CMO, said that fitting a five‑year maturity timeline into one year creates a major operational challenge and that security should be designed for machine speed from the start. SailPoint is enhancing its Agentic Fabric to locate undisclosed AI agents and to supply runtime authorization that can instantly shut down an agent.

The New Agent Audit feature converts agent inventory and governance data into framework‑aligned reports that they can export, tailor, or schedule. Agentic Fabric provides real‑time monitoring that applies policy‑based redaction and blocking and detects behavioral risks. Human Fabric will provide next‑generation access requests that include an embedded Harbor Pilot Access Request Agent.

A forthcoming Harbor Pilot Policy Agent will allow administrators to describe and create access policies for humans and AI agents using plain language. Agentic Fabric offers single‑click controls for managing agent lifecycles and includes a feature to end agents instantly. The integration of Entro Security into SailPoint Agentic Fabric will add credential lineage, exposed secret discovery and prompt‑intent monitoring.