started · updated
KubeCon North America 2026 addresses AI agent identity and policy management
The KubeCon + CloudNativeCon North America 2026 conference in Salt Lake City has highlighted critical challenges regarding AI agent identity and infrastructure security. A primary concern discussed is the difficulty of assigning stable cryptographic identities to AI agents within Kubernetes clusters. Unlike traditional workloads, AI agents exhibit non-deterministic, context-dependent behavior that complicates existing standards like SPIFFE and SPIRE. Data indicates a growing trend in this area, with reports suggesting that 72 percent of organizations now manage more non-human identities than human ones.
In addition to identity management, the conference featured practical applications of the Kyverno policy engine. Engineers from AT&T demonstrated how they use Kyverno with CEL policy types to implement attestation-driven admission control. This process allows for more granular deployment decisions, such as denying images in production environments if they lack specific signed SBOMs or sufficient code reviews, even if the image signature is valid. Benchmarks presented during the event showed significant performance improvements in Kyverno admission latency and throughput in recent versions.
Entities
AT&T · KubeCon + CloudNativeCon North America · Kyverno · SPIFFE · Salt Lake City