IEEE Cloud Summit 2026: The Tunnels No One Mapped
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
A recap of IEEE Cloud Summit 2026 highlighting key security and architecture themes from agentic AI deployments. Salesforce shared how they made an autonomous Kubernetes right-sizing agent production-ready by separating LLM reasoning from deterministic computation via an Integer Linear Programming solver. AWS's Dr. Chae Clark demonstrated context-aware security for agents, applying least privilege to the context window itself to prevent data leakage. A forensics researcher presented AgentTrace, an OpenTelemetry-based approach to making agent actions replayable and auditable, and highlighted real attacks including memory injection against Amazon Bedrock and EchoLeak against Microsoft Copilot. Capital One's GuardOn project was introduced as a browser extension that enforces Kubernetes policy at the pull request review stage. The overarching themes: non-human identities are massively over-privileged, probabilistic systems should only handle ambiguous tasks, and forensic traceability must be a default design requirement for agentic systems.
