In my view, you should almost never send commands over safety, only permissives. A signal to stop running can get lost in the way, with the effects you describe. However, if you instead send a signal every 50ms that says it is OK to continue operating (for now), then when it no longer receives that signal, it must shut down automatically.
The safety program should do the SAFE thing, which is to assume the worst for any untransmitted data. In the systems I work on, there is a watchdog timer for every communication, and if any communications trip the timer, it assumes a predetermined substitute value for all of those signals, which almost always force it to stop. In multi-zoned systems this can get more complicated, but at least partial shutdowns would typically be required.
Once you have a proper communications scheme, the hard part with wireless safety is to get the wireless system reliable enough to guarantee uptime. The communications themselves guarantee the safety, but it is up to you to design a network that works often enough to be useful. As Firejo said, this can be a trick, especially in the 5GHz band. I've seen a lot of companies switching away from traditional enterprise wireless systems and towards industrial systems designed to minimize latency/jitter instead of maximizing bandwidth.
