Connecting a wallet is the first action almost every decentralized application requests. It is worth knowing exactly what that step grants and what it does not.
The connection shares an address
The site asks the wallet for permission to see which accounts it controls, and the wallet returns the addresses the user selects.
Nothing is signed and no fee is paid. The result is that the site knows an address and can read everything that address has ever done on chain.
That reading is public information anyway, but the connection links it to a browsing session, which is a meaningful loss of separation between activity and identity.
Connection is not spending permission
A connected site cannot move funds. Every state-changing action requires a separate signature that the wallet presents for approval.
The confusion arises because connecting and signing look similar in the interface, and a site can request a signature immediately after connecting.
Users who approve prompts reflexively lose the protection this separation was designed to provide, since the second prompt is the one that matters.
How the wallet knows which network
The site can request a switch to a specific chain, and the wallet asks the user to confirm it, including adding networks it does not already know.
Adding a network means accepting a node endpoint the site supplied, which then becomes the source of the balances and transaction results the wallet displays.
A hostile endpoint cannot forge signatures but can misreport state, which is enough to mislead someone into approving a transaction they misunderstand.
Connections persist longer than expected
Most wallets remember a connection, so returning to a site restores access without a new prompt, often long after the user last thought about it.
The permission list accumulates quietly, and few users review it, leaving dozens of sites able to read the address whenever the browser visits them.
Clearing connections is straightforward in wallet settings, and unlike revoking a token approval it costs nothing because it never touched the chain.
The separation worth maintaining
Using a separate address for experimentation limits what any connection reveals, since the exploratory address has no history worth linking.
Holdings kept in an address that never connects to anything cannot be exposed by a signature prompt, because no site ever gets to present one.
This is the practical version of the security advice, and it works because the attack surface is the act of approving, not the act of holding.