To claim that the actual semantics of a word can be defined by anything other than the behavioural tendencies of its users is, at best, highly controversial.

On the contrary, it's trivially true. If semantics depended exclusively on behavior patterns, then novel thoughts would not be expressible. The meaning of the word "yellow" does not logically depend solely on which yellow objects in the universe accidentally happen to have been labeled "yellow" by humans. It is entirely possible that, sitting on a planet somewhere in the Andromeda galaxy, is a yellow glekdoftx. Under the negation-of-my-theory (I'll try not to strawman you by saying "under your theory"), that would be impossible, because, due to the fact that humans have never previously described a glekdoftx as "yellow", the extension of that term does not include any glekdoftxes. Examples like this should suffice to demonstrate that semantic information does not just contain information about verbal behavior; it also contains information about logical relationships.

edible" actually means "safe to eat

Guess what: I agree! Here, indeed, is my proof of this fact:

1. "Edible" means "able to be eaten".
2. In the relevant contexts, "able to be eaten" means "safe to eat".
3. Therefore, "edible" means "safe to eat".

See how easy that was? And yet, here I am, dealing with a combinatorial explosion of hostile comments (and even downvotes), all because I dared to make a mildly nontrivial, ever-so-slightly inferentially distant point!

Insert exclamation of frustration here.

Words mean exactly what people use them to mean - there is no higher authority

Yes, that thought is in my cache too. It doesn't address my point, which is more subtle.