Enterprises rush to deploy AI agents, straining identity controls

New research from Ping Identity finds enterprises are deploying autonomous AI agents faster than identity controls and governance can handle. The firm reports identity systems built for human users are being forced to operate continuously, creating gaps in visibility, accountability and authorization at the moment actions are executed.

The Ping Identity report says agentic AI can combine individually legitimate permissions in unexpected ways, producing actions that fall outside existing safeguards. Researchers identified delegation that lacks visibility and ‘‘sub-agent spawning,’’ where an agent creates other agents or chains of activity, which can make operations untraceable and audit trails incomplete.

The report adds that agents are bypassing human decision-makers in patterns not anticipated by OAuth and OIDC frameworks. Where systems do not continuously re-evaluate authorization, context can leak between systems, raising questions about permission inheritance, liability and enforcement in agent-to-agent interactions.

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

The report cites external research on related risks. IBM’s 2025 Cost of a Data Breach report found 13% of organizations reported AI-related security breaches and 97% lack adequate access controls for AI systems. Separate research from SANS found non-human and AI identities such as service accounts, API keys, automation bots and workload identities are multiplying faster than organizations can secure.

More than three-quarters of organizations surveyed by SANS said they were seeing growth in non-human identities while governance failed to keep pace. Of those already using AI agents that require credentials, 5% of security leaders reported they did not know whether agentic AI was running in their environments.

“These trends reflect a broader shift in identity requirements,” observed Martin Kuppinger, founder of KuppingerCole Analysts. “Organizations will need to extend identity and authorization models to preserve control, accountability and trust as environments grow more dynamic.”

The Ping Identity report recommends moving to continuous authorization checks, improving visibility into delegated credentials and strengthening governance for non-human identities. The firm advises applying identity controls at the moment actions occur and monitoring agent-driven workflows on an ongoing basis.