Here is an example using Nami and blockfrost for you to review:
Another smaller example:
In general, all wallet content will be viewable once the user allows for a website to 'connect'. Sending assets will obviously require the user to enter the wallet spending password which is used to ultimately sign the transaction.
('wallet content' above isn't quite right as really it's assets and transactions from the actual blockchain for this wallet address).
My understanding is that wallet interaction is all on the client side. The wallet code gets injected into the page during/after loading.
So nothing sensitive needs to be sent back to your web server and you do not need to handle the spending password. You could choose to send back to your web server whatever information that you want from the wallet but this is already publicly available information for that particular wallet address.
So if a wallet has allowed a website to connect and that website is controlled by a bad actor they could try to initiate a sending out of assets and the wallet will automatically prompt the user for their spending password (this is built into the wallet and out of control of you or the bad actor).
If the user is not paying much attention they might just enter their password without thinking about it - hard to know what percentage of people would do this.
Is their a particular scenario that you have in mind?