Barak Sternberg Joins the Execweb Podcast to Talk AI Agent Security

Barak Sternberg, CEO and Co-founder of Tenet Security, joined Episode 2 of the Execweb Cybersecurity GTM Podcast, hosted by Val Tsanev of Execweb (Acquired by CyberRisk Alliance). In the episode they discuss one of the most underexplored problems in enterprise security: the attack surface autonomous AI agents introduce, and why the industry's current defenses aren't built for it.

A New Attack Surface, Not a New Vulnerability

Autonomous agents don't just run code: they reason through context, call tools, and take sequences of actions that were never individually pre-approved. That design is also what makes them a difficult target to defend. The threat isn't just a misconfigured permission or a vulnerable library. It's the reasoning process itself, operating in real time, in ways that no static policy or pre-deployment scan can anticipate.

The conversation covers agentjacking, attacks in which adversaries manipulate an agent's reasoning context through prompt injection or poisoned tool outputs, causing the agent to act on the attacker's behalf while appearing to operate normally. The challenge isn't just that these attacks are hard to prevent. It's that the tools enterprises rely on today were never designed to detect them. EDRs and firewalls read code execution and network traffic. They don't interpret intent. An agent that has been manipulated into exfiltrating data through a sequence of individually permitted API calls won't trigger a single alert.

For a full breakdown of how agentjacking works at the mechanics level, see What is Agentjacking? on the Tenet blog.

Why This Is an Immediate Risk

One of the clearest signals from the episode is that CISOs are already treating this as an operational concern, not a future one. AI agent deployments are in production now: across customer service, software development, security operations, and financial workflows. The question has shifted from whether to secure them to what securing them actually requires. Barak's position is consistent: visibility at the reasoning layer, continuous rather than periodic, focused on what an agent is doing and intending rather than what infrastructure it touches.

The full conversation is available now. 

If you're working through how to establish visibility into your AI agent layer, reach out - we're happy to share what we're seeing in the field.

‍