Grab’s security team has built Palana, a Kubernetes-native secure execution platform for autonomous AI agents. InfoQ reports that the platform is designed for model-driven systems whose tool use, code writing, and prompt-injection risks are less predictable than deterministic software.
Palana contains those risks at the infrastructure level with isolated namespaces, out-of-process control planes, proxy-mediated access, and Vault-backed secrets. That design treats agent security as an execution-environment problem, not only a prompt-policy problem.
The project is a strong example of how enterprises are adapting cloud-native controls for agentic workloads.