October 15, 2026 -
Hardware-based identity and attestation are critical to protecting credentials and workloads as AI agents gain autonomy across data centers and edge devices.
AI agents that book travel, send emails, execute code and access connected services on behalf of users introduce security risks beyond those associated with conventional generative AI. PUFsecurity Corp., a subsidiary of eMemory Technology Inc., argues that protecting these systems requires hardware Roots of Trust embedded in the chips running them.
In its technical perspective, Building Trusted AI Agents from the Silicon Root of Trust, the company examines how hardware-based identity, secure key management and remote attestation can help establish trust in increasingly autonomous AI systems.
Unlike chatbots, which generate content for people to evaluate, AI agents can act on their outputs using delegated credentials and access privileges. A compromised agent could therefore execute unauthorized actions before a user has an opportunity to intervene.
Software isolation remains an important defense. Meta’s Muse architecture, for example, runs each user’s agent in a dedicated cloud virtual machine (VM), with an isolated runtime, file system and network interface. A separate Sentinel agent controls network access and connector actions, while surrogate tokens prevent the runtime from directly accessing real credentials. Credentials are exchanged at the network boundary after requests are approved.
However, Meta states that its launch Secure VM does not prevent the company from accessing data when necessary to support, secure or operate the service. It plans to introduce a Muse Confidential VM, in which the entire VM is encrypted using a key held only by the user.
PUFsecurity notes that such protection depends on the underlying hardware. A platform must be able to demonstrate that it is genuine and has booted the expected software before confidential data or cryptographic keys can be released.