I have no idea how AWS defines it, but to my mind it's about two things: intent and impact. Intent: What is the connection's purpose? To use a service, to map

I have no idea how AWS defines it, but to my mind it's about two things: intent and impact. Intent: What is the connection's purpose? To use a service, to map a topology, to identify potentially vulnerable targets, to consume resources? Impact: How does the act of connecting (once or many times) affect the remote end of the connection? Is a critical resource being exhausted (network connections, cpu, memory, etc)? Is the remote end's service still available for its intended purpose? There's no magic spot on the scales but the further you get away from "to use the service" on the Intent scale and "no detectable impact" on the Impact scale, the more trouble you're likely to run into.