In this episode of the Cybersecurity Awesomeness Podcast, host Chris Steffen is joined by Alex Salazar, founder and CEO of Arcade.dev, to explore the critical intersection of Artificial Intelligence, security, and agentic workflows. As AI transitions from a passive search tool to active "agents" capable of executing real-world tasks—such as modifying code, managing workflows, or handling data—traditional security paradigms are severely tested.
Salazar emphasizes a vital distinction: model guardrails (teaching an AI "character" and ethics) are not enough, just as raising a well-behaved child doesn't mean handing them the keys to a bank account. True security requires robust governance, delegated authority, and runtime permissions. Rather than giving autonomous agents direct, unmitigated access to systems, Salazar advocates for deterministic, runtime security layers that evaluate every requested action before execution. Ultimately, the guests agree that while agentic AI introduces novel risks, the industry will adapt through better authorization frameworks, much like previous shifts to cloud and virtualization.