Security, Trust & ComplianceTools-in-Action30min
Building the Trusted Agent Ecosystem
The session presents a Trusted Agent Ecosystem framework to address user trust barriers in the Agentic Web. Using Confidential Computing and Binary Transparency, it secures tools, inference, and remote agents via verifiable WebAssembly sandboxes and Confidential GPUs, enabling developers to build trusted agentic services.
talk.summaryAiDisclaimer
Ivan PetrovGoogle DeepMind
Patrick McGrathGoogle DeepMind
talkDetail.whenAndWhere
Monday, October 5, 18:20-18:50
TBA 2
talks.roomOccupancytalks.noOccupancyInfo
The industry is rapidly introducing agentic services that are accelerating the arrival of the "Agentic Web". Yet adoption is still bottlenecked by the lack of user trust. Users are understandably afraid to grant remote AI agents access to their private data or authorization tokens. To overcome this, we need robust solutions that empower all developers to create agents that users can trust.
In this session, we introduce a comprehensive framework for building the Trusted Agent Ecosystem. By leveraging Confidential Computing and Binary Transparency, we provide verifiable trust across the entire AI stack: securing MCP tools, model inference and remote agents connected via A2A. We will present our open-source implementation of the framework that leverages verifiable WebAssembly sandboxes and Confidential GPUs to build trusted agentic services.
In this session, we introduce a comprehensive framework for building the Trusted Agent Ecosystem. By leveraging Confidential Computing and Binary Transparency, we provide verifiable trust across the entire AI stack: securing MCP tools, model inference and remote agents connected via A2A. We will present our open-source implementation of the framework that leverages verifiable WebAssembly sandboxes and Confidential GPUs to build trusted agentic services.
Ivan Petrov
Ivan Petrov is a Senior Software Engineer at Google DeepMind who is working on security and privacy for the agentic ecosystem. He also holds a PhD from Moscow State University, where he focused on intrusion detection systems and software-defined networking.
Patrick McGrath
Patrick McGrath is a Senior Software Engineer at Google DeepMind specializing in agentic security and privacy for next-generation AI systems. With a strong foundation in server-side architecture, he has engineered critical infrastructure for high-scale platforms, including Private AI Compute and Google Maps Location services. Over the past five years at Google, his work has focused on building provably private solutions that safeguard user data at a global scale.