What Breaks When the Model Is Fooled: Securing Agentic AI and Skills
An in-depth security briefing that guides developers and security leaders on moving beyond probabilistic model alignment to establish multi-layered, deterministic controls over autonomous agents and their modular skill ecosystems
π§ Listen to this Episode
Show Notes
As AI deployments transition from basic chatbot interfaces into autonomous agents capable of executing complex real-world workflows, organizations are discovering that traditional input filtering can no longer guarantee safety. This podcast explores the critical security boundaries mapped across OWASP's main AI safety directories: the GenAI LLM Top 10, the Agentic Applications Top 10, and the newly released Agentic Skills Top 10. Join us as we dissect core vulnerabilities like malicious skills, over-privileged credentials, and supply chain compromises, outlining how security teams can build robust, system-level boundaries to control the blast radius when a model is inevitably fooled.
Β
Sponsors:
Share this episode
Enjoying CISO Insights?
Subscribe to get new episodes delivered directly to your podcast app.
Related Episodes
Securing the AI Supply Chain: The G7 SBOM Guidelines
This episode unpacks the G7βs framework for an AI Software Bill of Materials (SBOM), detailing how documenting system components across seven key clusters helps organizations track vulnerabilities and...
βΆοΈ Listen Now
The Mythos Paradox: Leaks, Lawsuits, and the AI IPO of the Century
Explore how Anthropicβs rollout of the highly dangerous Claude Mythos model, combined with accidental source code leaks and a Pentagon standoff, is shaping the narrative for a historic $60 billion IPO...
βΆοΈ Listen Now
The 40-Minute Collapse: How Fake Compliance Broke the AI Supply Chain
A devastating supply chain attack on the AI startup Mercor has exposed a web of fabricated security certifications and invasive workplace surveillance, triggering a massive data theft and unprecedente...
βΆοΈ Listen Now