Transcript
employees are increasingly operating beyond the guardrails. That's just one of the findings in Okta's new report, AI Agents at Work 2026, a global survey of nearly 800 business leaders and employees across seven countries, revealing the growing security, identity, and governance risks of AI agents in an enterprise. First, shadow AI is widespread. The vast majority of executives surveyed, 92%, say autonomous AI agents are already being used across their organizations. And while 90% of executives say they're confident they have visibility into those AI tools, more than half of employees admit they're using AI tools without approval, often through personal accounts. And why? Well, 57% say approval processes are too slow or difficult. And they say approved tools don't meet their needs. Second, governance gaps are already leading to real-world consequences. More than half of executives surveyed, 58%, say their organizations experienced an AI-related security incident or close call within the last year. Yet only 53% say their organizations have an established strategy guiding AI deployment. Third, employees are sharing sensitive information with AI tools. 29% of workers admit to sharing confidential company documents with AI systems. 36% are sharing HR-related information. And 16% report sharing login credentials and passwords. Organizations are also increasingly granting AI tools access to internal systems like inboxes, collaboration platforms, cloud storage, and CRM databases. Fourth, AI policies remain unclear across many organizations. While 65% of executives believe their company's AI usage policies are very clear, 57% of workers disagree, describing these policies as unclear, difficult to find, or non-existent. And finally, identity and access controls remain inconsistent for AI agents. Only 34% of organizations say they consistently apply the same security controls to AI agents as they do to human employees. That creates what the report describes as a dangerous double standard. Now that doesn't mean AI agents aren't driving real productivity gains. The concern is whether enterprise governance, visibility, and identity controls are evolving fast enough to keep up. So what can organizations do about it? The report recommends assuming shadow AI already exists and making AI discovery a priority. It also argues organizations need to make secure AI workflows easier to use because when approved systems create too much friction, employees will continue finding workarounds. And perhaps most importantly, the report says AI agents should be treated as first-class identities, with their own visibility, governance, and permissions. To learn more, check out our recent series exploring the secure agentic enterprise and the role of Okta's latest solution, Okta for AI Agents, in addressing these challenges. You can watch the full series now in Okta's newsroom. Till next time, I'm Diana Blass.
