In this episode of "Cybersecurity Under Pressure", we dive deep into the complex and rapidly evolving world of Artificial Intelligence cybersecurity. As Large Language Models (LLMs) evolve into autonomous "Agentic AI" capable of interacting with environments and executing real-world actions, the attack surface—and the pressure on security teams—has never been greater.

Join us as we unpack critical lessons from real-world vulnerabilities, explore how threat actors are actively compromising these advanced systems, and break down what the new wave of European regulations means for the future of AI innovation.

Key topics covered in this episode:

The Anatomy of LLM Attacks: Discover why "black-box" tactics based on iterative searches (like the TAP attack) are proving faster and more effective at deceiving AI agents than complex "white-box" mathematical methods (like GCG).

The Invisible Threat of Indirect Prompt Injection (IPI): Learn how attackers hide malicious instructions in web pages, emails, and resumes—sometimes using white text on a white background—to hijack AI systems and exfiltrate sensitive data without triggering traditional defenses.

The Risks of Agentic AI: We discuss how giving AI memory, tools, and autonomy exposes organizations to new dangers, including model leakage (silent extraction of internal context) and feedback loops that amplify biases and errors.

Building Robust Defenses with MLSecOps: We explore the essential transition from traditional DevOps to MLSecOps. Get a practical guide on securing the entire machine learning supply chain—from data engineering to model monitoring—applying a "security by design" approach.

Navigating Regulatory Pressure (EU AI Act): We break down the strict requirements and heavy penalties under the European Union's AI Act for systems classified as "High-Risk", such as those used in critical infrastructure, hiring, education, and law enforcement.

Tune in to learn from these real-world threats and discover how to secure AI innovation before it's too late!