Coursera Spring Sale
40% Off Coursera Plus Annual!
Grab it
Explore the emerging security risks of AI autonomy in this conference talk that examines the Model Context Protocol (MCP) and its implications for cybersecurity. Discover how MCP enables large language models to autonomously chain tools and make decisions without human oversight, fundamentally changing the relationship between AI reasoning and execution. Learn about the dangerous scenarios that emerge when AI systems gain the ability to browse the web, write and execute code, and interact with various tools independently. Understand how this autonomy creates new attack vectors where prompt injection exploits and rogue plugins can cause cascading security failures, turning innocent outputs into malicious inputs that trigger unintended command sequences. Examine real-world examples and hypothetical scenarios of AI agents operating beyond intended boundaries in coding assistants, autonomous agent frameworks like AutoGen and CrewAI, OpenAI Agent SDK, and AI desktop environments. Gain insights into the critical need for rethinking trust boundaries and sandboxing strategies when AI systems themselves are making autonomous decisions, and discover why traditional security models may be insufficient for this new paradigm of AI-driven automation.
