As artificial intelligence shifts from answering questions to taking actions on our behalf, a difficult question follows close behind: how do you trust an autonomous agent with sensitive data? In 2026, the industry’s answer is increasingly built on confidential computing.
The concept being championed is what experts call a “trust layer.” At the Confidential Computing Summit 2026, a central theme is the necessity of building verifiable trust infrastructure to support the next generation of AI agents and sensitive model data. The argument is that as AI systems gain more autonomy and handle more confidential information, simply trusting that a cloud provider or platform will keep data safe is no longer enough that trust needs to be technically verifiable. Industry leaders are framing this as the next major bottleneck. One Microsoft product manager’s summit session is titled around the idea that trust is the next bottleneck and that the agentic economy needs confidential computing, while a Samsung Research team is exploring how to optimize on-device confidential computing for AI. The common thread is moving from assuming systems are secure to proving it, what one industry session described as a shift from trust assumptions to trust evidence.
That principle is converging with established security infrastructure. A DigiCert executive’s summit talk explores why public key infrastructure and confidential computing are coming together, while Amazon is presenting on building trust through secure cloud infrastructure with its Nitro system.
The stakes explain the urgency. Confidential computing is now reaching the mainstream, with use spreading across domains from AI to pharmaceuticals and finance, becoming a driver for collaboration, privacy, and regulatory compliance globally. For sectors handling highly sensitive data as patient records, financial transactions, proprietary models, the ability to run AI workloads without ever exposing the underlying data in plaintext is transformative.
The technology is even reaching beyond the cloud. A recently unveiled humanoid robotics platform builds in cybersecurity features including secure boot and confidential computing, signaling that the trust layer is extending into physical AI systems and not just data centers.
The broader message for organizations is that confidential AI is no longer a theoretical safeguard. As autonomous agents become woven into business operations, the infrastructure that lets them work with sensitive data ”provably and securely” is becoming a prerequisite rather than a luxury. The companies building that trust layer now are positioning themselves at the foundation of how AI will operate at scale.
