cyber.hendryadrian.com
Security News · darkreading.com

Agentic AI Raises New Challenges for Confidential Computing

· Original article ↗

Summary

Enterprise adoption of confidential computing is growing, but AI agents’ persistent context and memory create new risks. Google proposes per-agent encryption keys and crypto-shredding, while Apple and ServiceNow describe confidential-computing deployments.

Key points

  • Microsoft Azure CTO Mark Russinovich said progress on cost, speed, and workload complexity has helped confidential computing gain enterprise interest.
  • Google Cloud’s Nelly Porter warned that AI agents may retain sensitive information in their context and active execution state.
  • Porter proposed dedicated encryption keys for each agent’s memory and retained data; destroying a key could delete protected data through crypto-shredding.
  • Apple’s Private Compute Cloud architecture reportedly combines Google Cloud, Nvidia Blackwell GPUs, Intel TDX CPUs, and Google’s Titan security chip, with at least two independent hardware roots of trust.
  • ServiceNow uses secure enclaves on Nvidia H100 GPUs with Opaque Systems and Microsoft Azure to answer sensitive sales-compensation inquiries in about eight seconds, versus a previous four-day ticket process.

Article Details

Event Type
Industry developments in confidential computing for agentic AI, including proposed protections and enterprise deployments.
Impact
AI agents’ retention of context and memory may expose sensitive enterprise data beyond what current confidential-computing protections were designed to address. Google Cloud proposed per-agent encryption keys and crypto-shredding; Apple’s Private Compute Cloud and ServiceNow’s internal sales inquiry system illustrate deployments using confidential computing.

People

Dion HarrisNvidia senior director of high-performance and AI factory solutions; said Nvidia hardware is ready for agentic AI confidential computing.Kellie RomackServiceNow chief digital information officer; described the company’s confidential-computing system for answering sales employees’ sensitive commission-related inquiries.Mark RussinovichMicrosoft Azure chief technology officer and deputy chief information security officer; discussed confidential-computing progress and risks associated with AI adoption.Nelly PorterGoogle Cloud director of product management for confidential computing and encryption; described security challenges from agent memory and proposed per-agent encryption keys and crypto-shredding.Sanchit Vir GogiaGreyhound Research CEO who commented on Apple’s confidential-computing architecture.

Vendors

AppleBut as AI agents retain context and memory, companies are now facing new security challenges that are forcing a rethink of how confidential computing should protect AI “minds.”Google CloudGoogle Cloud proposes per-agent encryption keys and crypto-shredding.Intelinvolves Google Cloud, Nvidia’s confidential computing on Blackwell graphics processing units (GPUs), Intel central processing units (CPUs) with trust domain extension (TDX), and Google’s Titan security chip.Microsoft Azuresaid Mark Russinovich, chief technology officer and the deputy chief information security officer at Microsoft Azure, at the event.NvidiaBut as AI agents retain context and memory, companies are now facing new security challenges that are forcing a rethink of how confidential computing should protect AI “minds.”Opaque SystemsThe technology runs on Nvidia’s H100 GPUs, Opaque Systems’ confidential AI computing framework, and Microsoft’s Azure Cloud.

Products

Blackwellit first supported confidential computing only on single GPUs before extending it across Nvidia’s newer Blackwell chips, closing the performance gap that had held the technology back,” Russinovich said.Intel TDXgraphics processing units (GPUs), Intel central processing units (CPUs) with trust domain extension (TDX), and Google’s Titan security chip.Nvidia H100The technology runs on Nvidia’s H100 GPUs, Opaque Systems’ confidential AI computing framework, and Microsoft’s Azure Cloud.Private Compute CloudApple’s Private Compute Cloud uses a multilayer confidential AI architecture.Titan security chip(GPUs), Intel central processing units (CPUs) with trust domain extension (TDX), and Google’s Titan security chip.

Related Articles