Modern Java profilers face a fundamental tradeoff: JFR's ExecutionSample event is safe but biased, while AsyncGetCallTrace provides accuracy but relies on unsupported JVM internals. Engineers from Datadog, SAP, Amazon, and the OpenJDK community collaborated to solve this by introducing a new first-class CPU profiling event in JDK 25. The solution uses cooperative stack walking — recording lightweight execution points and deferring stack unwinding to safepoints — eliminating the need to choose between safety and accuracy. The new JFR-based CPU sampling event is available as an experimental feature in JDK 25, and the Datadog Java profiler already uses it when available. Future work includes native frame support, on-demand stack capture, and eventually eliminating AsyncGetCallTrace dependency for wall-clock sampling.