Of course as usual the legacy media article fails to address that key issue. If I had to guess I'd say that their system uses some sort of HSM to sign chain transactions but there are not strong controls on what txn it will agree to sign. Therefore somewhere in their code there's a place where you can just insert your transaction du jour and it'll go through. Once you have that scenario anyone who can modify the system's code will be able to steal as much money as they want. So the "developer laptop" may have only been used to push a PR or perhaps change code via a more direct route -- publishing a new container image or other build artifact. Using AI tools I think it would be quite feasible to point an agent at the pile of code on that laptop and just ask it "what code do I need to change to allow any transaction to be sent on-chain"? So the attacker didn't need to have deep knowledge of the system.

Edit: Binance has now published some details on the attack. The attacker subverted an EIP-7702 contract rather than as I theorized an HSM authorization process.