AI Security: From a Threat Researcher’s Perspective

AI security has become a catchall term - models, prompts, data, infrastructure, agents - but the real risk driving every AI incident comes down to one thing: access. 

Today’s AI agents don’t just generate text. They run workflows, execute commands, make API calls, and operate inside your environment using real credentials. And here’s the uncomfortable truth: most organizations cannot answer the three questions that matter most: What can your agents' access? What can they do? And who’s governing any of it? 

In this session, a BeyondTrust Phantom Labs™ threat researcher broke down why agentic AI isn’t a brand-new security domain at all - it’s an accelerant poured onto a longstanding problem: identity sprawl and uncontrolled access. 

Drawing on active research and real-world testing, we explored: 

• How agentic AI explodes the attack surface by adding autonomous actors to already fragile identity systems 

• Why most AI deployments inherit access by default—with no visibility, boundaries, or safety checks 

• Why “best practices” for agent security don’t exist yet, and what’s realistically enforceable right now 

Walk away with a clear, practical framework for assessing agentic AI risk, why identity must become the control plane, and how to apply emerging guidance - even while the market plays catchup.