ERNW's Black Hat research examines isolation failures that remain possible when machine-learning notebooks, pipelines or tenant workloads share Kubernetes infrastructure. Namespace RBAC, network policies and pod restrictions provide a baseline, but shared controllers can turn a tenant-controlled resource into access elsewhere. The paper illustrates this through Kubeflow session handling, Istio routing and Traefik cross-namespace references. In particular, Istio confirms that a tenant allowed to create VirtualService resources can influence mesh-wide routing; it treats this as expected behavior requiring additional safeguards, not a vulnerability solved by an ordinary version upgrade. Destination authorization policies and mutual TLS still constrain forwarded requests. The authors' assessment method starts by finding explicit and hidden tenants, inventories shared components, and traces both configuration and runtime interactions across trust boundaries. Remediation combines applicable vendor fixes with restricted controller permissions and admission policies. For AI platforms, the important boundary is what a notebook or agent can cause shared infrastructure to do, not merely which namespace contains its pod.
Kubernetes multi-tenancy: test shared controllers beyond namespace RBAC
Related research
More curated notes connected through Agent Security and AI Red Teaming.
Pacing model development in an era of cyber-critical capabilities
OpenAI says preliminary evidence that Astra may meet its Critical cybersecurity threshold led it to pause frontier reinforcement-learning work for two weeks and keep its largest planned run on hold. New safeguards include stronger workload and network isolation, continuous boundary testing, token-level monitoring that escalates suspicious tool activity, and broader alignment checks for deception, reward hacking, and unauthorized access.
OWASP Top 10 for Agentic Applications for 2026
OWASP's community guide organizes agentic-system risk into ten categories, including goal hijacking, tool misuse, identity and privilege abuse, memory poisoning, insecure inter-agent communication, cascading failures, and rogue-agent behavior. It provides a shared taxonomy and mitigation starting point rather than a certification checklist or evidence that a deployed system is secure.
A blueprint for democratic governance of frontier AI
OpenAI proposes a three-part U.S. frontier-AI governance model: harmonize emerging state safety laws into a federal baseline, strengthen CAISI as an evaluation and standards institution, and coordinate a broader resilience program. Proposed controls include severe-risk evaluations, transparency reports, independent audits, safety-incident reporting, model-weight security, whistleblower protection, and periodic technical assessments.