“It’s going to become the No. 1 attack vector that we’re going to see,” he states. “To make our lives easier as humans, we’ve given [AI agents] the crown jewels to everything.”

Security Leaders Weigh In 

Ryan McCurdy, VP, Liquibase:

AI agents create a different kind of insider threat because they can be both the target of an attack and the thing taking action inside the enterprise.

If an attacker compromises an agent that already has legitimate credentials and access to critical systems, traditional access controls only get you so far. The agent may still be doing exactly what it has permission to do.

That’s why identity and permissions are only part of the answer. Organizations also need to govern what the agent can actually change. A compromised agent and a well-behaved agent should face the same policies and controls before their actions reach production.

The source of the change isn’t what determines risk. The change itself does. As agents gain more autonomy, governance has to follow the action through every tool, permission, and system it touches.

Kevin Surace, CEO, TokenCore

AI agents are like interns with root access and no fear of HR.

They can be compromised, manipulated, or impersonated, then operate at machine speed. Organizations must limit both their permissions and their blast radius.

For consequential actions, agents should require human approval tied to a trusted device and verified with a fingerprint. Approval from an email or popup or another agent is not acceptable. Otherwise, a compromised agent or attacker could authorize the very transaction the control was meant to stop.

The rule should be simple: agents can recommend, prepare, and execute routine work, but only an authenticated human can approve an irreversible or high consequence action.

Noelle Murata, Chief Operating Officer, Xcape, Inc.

Granting broad system access and operational autonomy to artificial intelligence agents transforms them into high-value, non-human insider threats capable of compromising enterprise networks. As threat actors pivot from manipulating model inputs to targeting agent identities, over-privileged tool integrations allow compromised agents to execute unauthorized application programming interface (API) actions, exfiltrate data, or move laterally across connected systems. Security leaders must apply zero trust principles to agentic workflows by replacing persistent access credentials with scoped, short-lived tokens, enforcing least-privilege tool permissions, and extending continuous identity monitoring to all non-human entities operating across the Internet.

Critical Takeaways:

• Autonomous artificial intelligence agents with broad permissions function as non-human insider threats, enabling lateral movement and unauthorized execution.
• Enterprise security controls must treat agent identities with zero trust rigor, substituting persistent keys with short-lived, scoped tokens.
• Security teams need continuous behavioral monitoring and strict application programming interface access boundaries to detect anomalous agent interactions.

An over-privileged agent with broad tool access is not an efficiency gain, it’s an unmonitored insider waiting to be exploited.