AI Agents Have Too Much Access. Here’s
The Fix.

AI agents are already taking real actions across your systems, like your CRM, your ERP or even your code base. Most companies are still treating  agents like human users. And that’s a massive security risk,

Traditional IAM was designed for people. One identity. Accountable  users. Time to review access. And, to be fair, human access is still a big challenge. 

Controlling access of AI agents is exponentially more difficult. Agents get connected to multiple tools, operate at machine speed and have zero accountability or liability. Today, we still give them and long-lived credentials with broad access.

It feels like: The world is moving towards an agentic workforce. But forgot about the underlying security model - that is required to adopt agents at scale. 

Join Johannes Keienburg, CEO and Founder of Cakewalk, and Herman Errico, Senior Product Manager at Vanta, for a practical session on how to control AI agent actions without slowing innovation.

Herman recently published a paper on “Autonomous Action Runtime Management” (AARM), which has received significant attention. AARM is a cross-industry solution to the problem: a runtime security specification that evaluates AI agent actions before execution using policy and session context, treating tool execution (not the model) as the primary security boundary.

What You’ll Learn

• Why AI agent risk is different from human access risk
• Why static permissions are risky 

• Why LLM guardrails fail
• What AARM is all about 
• What a dynamic (job-based) authorization model looks like in practice
  

This is a practical session for Security & IT teams as well as CTOs  who need to enable AI agents without losing control.

Even if you can't attend, sign up anyway and we'll send you the recording.