Kafka can probably guarantee exactly ones semantics on publishing (conditions apply). It definitely cannot guarantee exactly once semantics on the consumer and processing side. Imagine a scenario where you receive a message from Kafka and process it, but the processor crashes or has a network partition right after. There's no way for the message to be acknowledged and you either have to design your system to be idempotent or handle exactly-once semantics further down the stack.

Databases have been handling exactly-once semantics for decades now. What Kafka is doing is not new and actually gives you a false sense of security when it comes to these kinds of things.