Current events: Identifying webpages by tapping the electrical outlet

摘要: Computers plugged into power outlets leak identifiable information by drawing variable amounts of when performing different tasks. This work examines the extent to which this side channel leaks private about web browsing an observer taking measurements at outlet. Using direct AC consumption with instrumented outlet, we construct a classifier that correctly identifies unlabeled traces webpage activity from set 51 candidates 99% precision and recall. The rejects samples 441 pages outside corpus false-positive rate less than 2%. It is also robust number variations in loading conditions, including encryption. When trained on two computers same webpage, labels further either computer. We identify several reasons for consistently recognizable consumption, system calls, propose countermeasures limit leakage information. Characterizing may help lead practical protect user privacy untrustworthy infrastructure.