Securing AI Agents at Runtime
Techstrong TV Interviews
•
17m
AI agent runtime security is quickly becoming the frontier problem for modern security teams. David Mytton, founder and CEO of Arcjet, joins Alan Shimel on Techstrong TV. Furthermore, he explains why every action inside an agent workflow now needs its own control.
About David Mytton
David is a third-time founder who started in cloud monitoring in 2009 and sold that company in 2018. In addition, he ran product engineering at StackPath and later launched the console.dev DevTools newsletter. Consequently, he brings a rare mix of operator, product and developer community experience.
Why AI agent runtime security matters
David explains that Arcjet works in three stages, observe, identify and then enforce. Meanwhile, the platform ingests OpenTelemetry data and adds an SDK inside the agent code path. Therefore, teams get AI agent runtime security on every tool call, not just at the edge.
He walks through prompt injection detection, PII detection and tool guards as core controls. Furthermore, pricing scales with observed applications and enforced policy usage. As a result, AI agent runtime security becomes a granular per action decision.
Behavior drift and the OpenAI Hugging Face incident
David revisits the OpenAI and Hugging Face incident as a failure of observability, not just sandboxing. In addition, agents drifted in behavior over months before the attack window opened. Consequently, AI agent runtime security has to catch slow drift and not only point in time abuse.
He references the Google Beyond Zero paper on behavior drift as a north star for the category. Meanwhile, most CISOs are still mapping which agents even exist in production. Therefore, guardian agent style controls remain early and unevenly deployed.
Where the agent security market is heading
David compares the current agentic security market to the early two thousands, with many vendors and few standards. Furthermore, Black Hat this year was a blur of agentic messaging without clear differentiation. Consequently, buyers should pick tools that treat every agent action as a policy decision.
Explore more artificial intelligence coverage and the latest Techstrong TV interviews.
For more information please visit arcjet.com
Up Next in Techstrong TV Interviews
-
Knowledge Engine Built for AI Agents
The knowledge engine for agents is quickly becoming a required layer for enterprise AI. Aaron Kao, VP of marketing at Pinecone, joins Alan Shimel on Techstrong TV. Furthermore, he explains why agents do not have a model problem, they have a knowledge problem.
About Aaron Kao
Aaron leads mar...
-
Cyber Ranges Help Municipalities Defe...
Critical Infrastructure Cybersecurity Needs Practical Training
Critical infrastructure cybersecurity is becoming more urgent as municipalities face more connected operational systems and more aggressive threat actors. In this Techstrong TV interview, Mike Vizard talks with Lee Rossey, CTO and co...
-
Stopping Rogue AI Agents in Real Time
Rogue AI agent detection is quickly becoming a top security concern. Naor Paz, CEO and co-founder of Capsule Security, joins Alan Shimel on Techstrong TV. Furthermore, he explains why agents demand a runtime approach and not just posture management.
About Naor Paz
Naor grew up around comput...