70% of Companies Now Let AI Make Human Decisions — but Half Can't Audit What It Does

Key takeaway: A new global report from Illumio finds 70% of organizations now let AI make decisions that required human approval a year ago — yet only 49% can see and audit every AI action. With 76% reporting an AI-related security issue in the past year, the gap between AI’s authority and oversight is becoming a serious security problem.

What the report found

Illumio, a breach-containment security company, published Licensed to Act: How AI Autonomy Is Outrunning the Limits Placed on It on October 21, 2026. The research was conducted by Sapio Research, which surveyed 1,100 IT and security decision-makers across 10 markets: the United States (200 respondents), plus the UK, France, Germany/Switzerland/Austria, the Middle East (UAE and Saudi Arabia), Japan, Australia, Singapore, Brazil, and Mexico (100 each).

The headline numbers, reported by Illumio:

70% of organizations now delegate decisions and actions that required human approval a year ago.
Only 49% can see and audit every action taken by an AI agent.
76% experienced at least one AI-related security issue or warning sign in the past 12 months.
35% say AI made a recommendation that would have created a security risk if followed.
26% have had an AI agent or tool connect to systems without security approval.
24% say AI accessed data it should not have reached, and the same share say AI produced or triggered an unauthorized action.

AI is gaining authority inside security teams themselves

The stakes look highest where AI is operating the defenses. According to the report, 67% of organizations use AI within cybersecurity operations. A third (32%) are comfortable letting AI run security investigations without human review, 29% are comfortable letting it deploy code changes unsupervised, and 26% say the same about granting or revoking access permissions.

Organizations are also explicitly trading risk for speed: 56% say they are willing to accept more cyber risk than they were 12 months ago. Among that group, over half permit autonomous AI actions in non-production environments, and 47% permit them in production environments.

Why it matters: AI as a new insider risk

Illumio frames the problem as a “new kind of insider risk.” Most AI security discussion has focused on how attackers might use the technology — including reports of autonomous AI agents being used to breach company sites — but this report looks inward: organizations’ own AI systems now hold credentials and connectivity to enterprise applications, data, and infrastructure. Unlike an external attacker that must first break into an environment, an organization’s own AI may already have what it needs to reach critical systems.

“The biggest AI risk may not be a system that breaks its rules. It may be one that was never given meaningful limits in the first place,” said Raghu Nandakumara, VP of Industry Strategy at Illumio. “AI agents are highly persistent in pursuing a goal, so trying to anticipate every action is a losing battle. Businesses must set strict limits on what they can access and do, then enforce those limits within the application and across the underlying infrastructure.”

The report’s central framing is a control gap: the authority organizations give AI is growing faster than their ability to audit and constrain it.

What organizations should actually do

The report’s practical recommendation is straightforward: limit what AI can reach, rather than trying to predict everything it might do. In practice that means:

Define permitted actions explicitly. Only 25% of respondents name explicitly defined permitted actions as one of their two main controls — a surprisingly small share given that 82% say AI agents require fundamentally different security controls than traditional software.
Apply Zero Trust to agents, not just people. Verify access continuously, grant least privilege, and contain workloads so an AI agent can only touch what it needs.
Make every AI action auditable. If only 49% can see and audit every AI action, logging agent decisions and actions is the immediate gap to close.
Keep humans in the loop for high-impact actions. Code deployments, access-permission changes, and incident response are the actions respondents are already letting AI take unsupervised — treat those as approval checkpoints first.

Analysis

The following is analysis, not reported fact. This survey arrives at a moment when AI agents are moving from advisors that recommend actions to operators that take them. The 70%/49% split is the number that matters most: permission is being granted faster than oversight is being built, and the 76% who saw AI-related issues last year suggests the consequences are already showing up. The finding that only 27% say frontier AI has made network segmentation or microsegmentation more important points to an awareness problem — teams know agents are different from normal software (82% say so) but are not yet treating their reach as the main thing to control. Expect enterprise AI governance to shift from “review the output” to “constrain the access” over the next year, because auditing everything an agent does is hard — limiting what it can touch is tractable.

Frequently asked questions

What did Illumio’s “Licensed to Act” report find?

The report, published October 21, 2026 and based on a Sapio Research survey of 1,100 IT and security decision-makers in 10 markets, found that 70% of organizations now delegate decisions that once required human approval to AI, while only 49% can see and audit every action their AI agents take. It also found that 76% experienced at least one AI-related security issue or warning sign in the past year.

Why is letting AI make decisions risky?

The risk is a control gap: AI is gaining authority faster than oversight is growing. According to the report, 35% of organizations say AI made a recommendation that would have created a security risk if followed, 26% had an AI agent connect to systems without security approval, and 24% say AI accessed data it should not have reached.

What is an AI insider risk?

It is the idea that an organization’s own AI systems pose a security risk similar to a trusted insider. Unlike an external attacker that must break in first, an internal AI agent may already hold credentials and connectivity to reach critical applications and data, so its mistakes or misuse can cause damage without any intrusion happening.

How can companies limit what AI agents can do?

The report recommends a Zero Trust approach: define permitted actions explicitly, grant AI agents least-privilege access, contain them with network segmentation so they can only reach what they need, and log every action so it is fully auditable. The key shift is from trying to predict everything AI might do to enforcing hard limits on what it can reach.

Who conducted the research and how?

Sapio Research conducted the study on behalf of Illumio, surveying 1,100 IT and security decision-makers across 10 markets: the US (200 respondents) and the UK, France, Germany/Switzerland/Austria, the Middle East, Japan, Australia, Singapore, Brazil, and Mexico (100 each). The figures cited are unweighted global averages.

Enjoying USABite’s coverage? You can add USABite as a preferred source in Google Search so our stories show up first.