SEND_SMS
Sends SMS silently, which turns the infected phone into a remote-controlled handset: confirming fraud, and proxying attacker messages through the victim's own number.
Runtime prompt, and the permission people most reliably misread. The instinct is premium-rate fraud, and that is a real use of SEND_SMS, but it is not what this kit does with it. The write-up is explicit that the role here is subtler.
What it actually buys is the ability for the implant to act as the victim. Three ways, in ascending order of severity: confirming fraudulent transactions so the victim's banking app sees an approval they did not give, relaying attacker-composed messages through the victim's own number, and issuing MMI or USSD-adjacent carrier commands.
The middle one deserves attention on its own, because it defeats a control rather than bypassing a prompt. Messages sent from the victim's number look, to every downstream system including the recipient and any fraud filter, like messages the victim sent. SMS proxying of this kind survives sender-ID checks precisely because there is nothing anomalous to detect. The victim's number becomes the attacker's sending identity.
Taken with CALL_PHONE the effect is a complete remote-controlled handset. It can place calls and send texts without any UI, so the operator can drive the victim's accounts from outside while every confirmation and every verification request lands on a phone the victim believes is idle.
Related: RECEIVE_SMS, READ_SMS, CALL_PHONE
Part of Few Android permissions I have seen used often. All permissions here were seen in the mprivhaan analysis.