Hajra Naz
Security teams are being advised to treat AI agents as identities in their own right, with credentials, scoped permissions and audit trails, because agents in finance, engineering, HR and customer operations increasingly act on live systems rather than merely suggest actions. The forecast from Token Security is that these identities will end up among the most privileged in the organisation, which makes permission scope and revocation the practical problem rather than model quality.
Token Security, an AI-focused cybersecurity firm, warns that AI agents will become the most privileged identities within large organizations in 2026. This shift could create new security and compliance risks as companies deploy these systems into live business operations, according to Security Brief United States.
Enterprise Adoption Across Key Functions
The company predicts that enterprises will roll out AI agents at scale across departments like finance, software development, HR, and customer operations. These agents will perform critical actions, access sensitive data, and trigger business workflows.
Token Security expects AI agents to outnumber human users in some areas, receiving broader permissions than traditional employees. This shift will require updated identity management strategies and regulatory frameworks that currently assume humans are the primary actors.
Read More: How AI Agents Are Changing Business Leadership and Everyday Decisions
The Need for Stronger Controls
Itamar Apelblat, CEO and co-founder of Token Security, emphasized that organizations will need clearer controls and accountability for AI agent activity. Meanwhile, CTO Ido Shlomo warned that reliance on static API keys and long-lived credentials could weaken agent security.
The firm anticipates that each employee may interact with multiple agents, which will shift identity risk from user accounts to the AI systems acting on their behalf.
Emerging Security Risks
Token Security predicts a rise in security incidents caused by:
Mismatches between human access rights and agent permissions
Misconfigurations introduced by AI coding tools
Agent identity failures that could drive future breaches
As AI agents gain more influence in enterprise systems, companies will need to rethink traditional security practices and ensure robust governance frameworks to prevent catastrophic failures.
How this prediction has aged
The direction looks right, the pace slower than the headline suggests. Agents are being deployed, but OpenAI’s own leadership has noted that enterprise AI has not yet reached core business processes, so for most companies the identity problem is arriving department by department rather than all at once. The planning advice holds either way, because the permissions quietly granted during a pilot are the ones nobody goes back to review.
