When a user wants to register a token for use across chains they first call InterchainTokenService::deployAndRegisterStandardizedToken on the local chain. This will use a user provided salt together with the msg.sender to create the tokenId which is used as the salt to create both the StandardizedToken and the TokenManager.
They can then use this to deploy their token to any chain that Axelar supports.
Relying on msg.sender across chain comes with some security considerations though. If the user/protocol dont control the address used as msg.sender across all chains that are supported by Axelar ITS, they are susceptible to the same hack that affected wintermute; where an old gnosis wallet was used that had an address that could be stolen on another chain.
If an attacker controls the msg.sender address on another chain, they can simply create a token and manager with the same salt that they control. This will give them the same tokenId. They can then send a message to the chain where the real token is and get funded real tokens. All theyve done is burn/lock their fake token on their sourceChain. 
I recommend this is highlighted as a risk in the documentation so third party protocols building on top of Axelar are aware of this risk. 
