AI agents are racing ahead of the security meant to govern them

SailPoint’s fifth annual Horizons of Identity Security report, titled “From frontier to fabric: Mastering the autonomous enterprise,” puts a number on a problem many security chiefs sense but few can measure. Based on a survey of 340 senior identity, IT, cybersecurity, and risk leaders across North America, Europe, Asia Pacific, and Latin America, the survey finds that 79% of organizations already run AI agents in production. Only 2% have deployed identity security tools built specifically to govern them.

The 40-to-1 mismatch

The report calls this the “AI velocity paradox.” Agents are querying databases and executing transactions, yet most are still managed with human-speed tools. About 85% of organizations rely on legacy or general-purpose identity systems, and AI agents already make up 22% of all non-human accounts. Just 4% run a fully modern identity stack, and 79% expect security vulnerabilities to rise.

A market stuck at the starting line

SailPoint grades companies on a five-tier “Horizons” scale, from no formal program to a self-governing, ecosystem-integrated identity fabric. Roughly 60% sit in the bottom two tiers, and fewer than 1% have reached the top.

The reason is a split between two maturities. Some 87% of leaders rate their human identity management as capable or better, but only 43% report mature processes for agent access. Specifically for agent identities, 54% sit at the lowest tier. That is worse than the 45% where human programs started in 2022. The report’s blunt conclusion is that the five-year human curve must become a one-year agent curve.

Confidence versus capability

Perhaps the most striking finding is an awareness gap. About 80% of respondents say the gap in their tooling is moderate or small, yet only 15% can provision machine access in real time. The same share has mature capability to detect shadow AI. Similarly, 57% feel confident about meeting regulations, but only 43% are prepared to produce evidence for an AI-related audit. Notably, 69% already include agents in audits, and 81% expect that workload to grow.

Fears are basic, not exotic

Leaders worry less about rogue AI than about old problems at new speed. Data leakage (35%) and excessive permissions (30%) top the risk list. As the report puts it, agents will behave exactly as configured, with too much access, for too long.

The hardest tasks are foundational: managing credential lifecycles (58%), preventing shadow AI (58%) and monitoring agent behavior in real time (57%).

Security that pays for itself

The report argues that securing non-human identities is a business accelerator. Some 62% report notable productivity gains, and 46% report safer, faster AI deployment. The most mature organizations are roughly twice as likely to see significant gains, and nearly three times as likely to report faster, safer AI rollout. Looking ahead, 76% expect better cyber insurance eligibility or terms.

The roadmap

SailPoint recommends three stages: discover every identity and shadow AI tool, govern them with standard lifecycles and clear ownership, and then protect them through zero-standing-privilege and automated threat response. The report cautions against rushing to advanced controls before the basics are in place. It also notes that 64% already govern agents through their existing central identity platforms.

A caveat

This is vendor research. SailPoint sells identity security, and the report favors a unified-platform approach. The top-tier sample is also tiny (just two organizations at Horizon 5), which the report itself flags as directional only. Still, the central message holds: identity, long treated as a human problem, is now a machine problem too.