AI agents are being deployed into production faster than enterprises can govern them, exposing gaps in identity systems designed for human users.

This is according to new research from Ping Identity; the firm’s report, From AI Agents to Trusted Digital Workers, highlights the challenges when governing AI agents and identifies a critical failure mode emerging in enterprise identity systems as these agents operate at runtime, beyond the limits of traditional access controls.

As organisations move AI agents into production environments, the focus is shifting from managing identity to controlling how identities act across systems, data, and workflows. Identity systems originally designed for human interaction are now being pushed to operate continuously, increasing pressure on existing models and exposing gaps in governance, visibility, and accountability at the moment decisions are executed.

“Enterprises are deploying autonomous AI faster than they can govern it,” said Andre Durand, CEO & Founder, Ping Identity. “Identity remains foundational, but in an agentic environment it must operate continuously. Control must be enforced at the moment an action occurs.”

Where Traditional Identity Models Break Down for AI Agents

The research describes a failure mode in which AI agents combine individually legitimate permissions in unintended ways, resulting in actions that bypass established controls and cannot be fully traced or governed. This failure mode represents a new class of identity risk in environments where AI agents operate autonomously across enterprise systems.

As the industry is quickly learning, access grants permission. It does not enforce control.

With AI adoption accelerating, organisations face new challenges, including delegation opacity and sub-agent spawning, where agent chains become untraceable and break auditability.

Further, implicit human assumptions in IAM, as OAuth and OIDC models rely on human decision-makers that agents bypass, creating risks.

Other challenges include context leakage across systems without continuous re-evaluation of authorisation, and new questions  around permission inheritance, liability, and enforcement in agent-to-agent interactions.

Recommended reading

• AI Agents Are Transforming Enterprise AI Adoption
• 65% of Orgs Reported AI Agent-related Incidents Last Year
• AI Agent Deployments Are Outpacing Security Posture
• AI Agents Go Mainstream, But Fragmentation Is a Major Obstacle

The Risk is Already Materialising

Independent research from KuppingerCole Analysts reinforce the urgency of governing AI-driven identity interactions, noting that AI agents already interact across enterprise identity systems while many IAM approaches remain focused on users and controlled environments.

The research also highlights measurable risk, citing findings from IBM’s 2025 Cost of a Data Breach report that show that 13% of organisations have experienced AI-related security breaches, and 97% of organisations lack adequate access controls for AI systems

Recent incidents, including enterprise data leaks and prompt injection attacks, demonstrate how gaps in AI governance are already being exploited in real-world environments.

Despite these risks, most identity and access management approaches remain centered on human users and static access decisions, leaving organisations unprepared to govern autonomous systems.