Samsung v Apple
Surprisingly, Samsung devices, which accounted for 23 per cent of the researcher's Android data set, show no evidence of implementing MAC address randomization.
Apple, meanwhile, introduced MAC address randomization in iOS 8, only to break it in iOS 10. While the researchers were evaluating devices last year, Apple launched iOS 10 and changed its network probe broadcasts to include a distinct Information Element (IE), data added to Wi-Fi management frames to extend the Wi-Fi protocol.
"Inexplicably the addition of an Apple vendor-specific IE was added to all transmitted probe requests," the paper explains. "This made identification of iOS 10 Apple devices trivial regardless of the use of MAC address randomization."
This shortcoming aside, Apple handles randomization correctly, in the sense that it properly randomizes the full 48-bits available for MAC addresses (with the exception of the Universal/Local bit, set to distinguish between global MAC addresses and the local ones used for randomization, and the Unicast/Multicast Bit).
The researchers find this interesting because the IEEE charges a fee for using the first three bytes of that space for CID prefixes, "meaning that Apple is freely making use of address space that other companies have paid for."
In a phone interview with The Register, Travis Mayberry, assistant professor at the US Naval Academy and one of the paper's co-authors, expressed surprise that something like 70 per cent of Android phones tested did not implement MAC address randomization.
"It's strange that Android was so vulnerable," he said. "It's just really bad at doing what it was supposed to do."
'Closest to being pretty good'
Apple, meanwhile, fared better in terms of effort, though not results. "Apple is the closest to being pretty good," Mayberry said, but noted that Apple devices, despite the advantage of hardware consistency, are still vulnerable to an RTS (Request to Send) attack. Sending RTS frames to an Apple phone forces the device to reveal its global unique MAC address, rather than the randomized one normally presented to the hotspot.
"No matter how hard you try, you can't defend against that because it's a property of the wireless chip itself," said Mayberry.
There was single Android phone that fared well. "The one Android phone that was resistant to our passive attacks was the CAT S60 which is some kind of 'tough' phone used on construction sites and the like," Mayberry explained in an email. "It did not have a recognizable fingerprint and did not ever transmit its global MAC except when associating. It was still vulnerable to our active RTS attack though, since like I said, that is a problem with the actual chips and effects every phone."
Mayberry was at a loss to explain why Apple shot itself in the foot by adding a trackable identifier to a system that previously worked well.
"I initially thought it might be to support some of the 'continuity' features where multiple apple devices can discover and exchange stuff like open browser tabs and clipboard contents but that came out in earlier versions of iOS," he said. "It also might be linked to the HomeKit features that they added in iOS to control IoT devices. Basically it would have to be to purposefully identify and discover other Apple devices that are not associated, otherwise we wouldn't see it in probe requests. All of this is pure speculation though and we really don't have a strong reason for it."
Mayberry said he hoped the research would help the industry understand the consequences of everyone doing things differently. There's no generally accepted way to handle MAC address randomization. "There are so many phones not using it," he said. "There should be a standard." ®