New SnailLoad Attack Exploits Network Latency To Spy On Users’ Web Activities

Longtime Slashdot reader Artem S. Tashkinov shares a report from The Hacker News: A group of security researchers from the Graz University of Technology have demonstrated a new side-channel attack known as SnailLoad that could be used to remotely infer a user’s web activity. “SnailLoad exploits a bottleneck present on all Internet connections,” the researchers said in a study released this week. “This bottleneck influences the latency of network packets, allowing an attacker to infer the current network activity on someone else’s Internet connection. An attacker can use this information to infer websites a user visits or videos a user watches.” A defining characteristic of the approach is that it obviates the need for carrying out an adversary-in-the-middle (AitM) attack or being in physical proximity to the Wi-Fi connection to sniff network traffic. Specifically, it entails tricking a target into loading a harmless asset (e.g., a file, an image, or an ad) from a threat actor-controlled server, which then exploits the victim’s network latency as a side channel to determine online activities on the victim system.

To perform such a fingerprinting attack and glean what video or a website a user might be watching or visiting, the attacker conducts a series of latency measurements of the victim’s network connection as the content is being downloaded from the server while they are browsing or viewing. It then involves a post-processing phase that employs a convolutional neural network (CNN) trained with traces from an identical network setup to make the inference with an accuracy of up to 98% for videos and 63% for websites. In other words, due to the network bottleneck on the victim’s side, the adversary can deduce the transmitted amount of data by measuring the packet round trip time (RTT). The RTT traces are unique per video and can be used to classify the video watched by the victim. The attack is so named because the attacking server transmits the file at a snail’s pace in order to monitor the connection latency over an extended period of time.

Read more of this story at Slashdot.

Nvidia Forecasted To Make $12 Billion Selling GPUs In China

Nvidia is expected to earn $12 billion from GPU sales to China in 2024, despite U.S. trade restrictions. Research firm SemiAnalysis says the GPU maker will ship over 1 million units of its new H20 model to the Chinese market, “with each one said to cost between $12,000 and $13,000 apiece,” reports The Register. From the report: This figure is said by SemiAnalysis to be nearly double what Huawei is likely to sell of its rival accelerator, the Ascend 910B, as reported by The Financial Times. If accurate, this would seem to contradict earlier reports that Nvidia had moved to cut the price of its products for the China market. This was because buyers were said to be opting instead for domestically made kit for accelerating AI workloads. The H20 GPU is understood to be the top performing model out of three Nvidia GPUs specially designed for the Chinese market to comply with rules introduced by the Biden administration last year that curb performance.

In contrast, Huawei’s Ascend 910B is claimed to have performance on a par with that of Nvidia’s A100 GPU. It is believed to be an in-house design manufactured by Chinese chipmaker SMIC using a 7nm process technology, unlike the older Ascend 910 product. If this forecast proves accurate, it will be a relief for Nvidia, which earlier disclosed that its sales in China delivered a “mid-single digit percentage” of revenue for its Q4 of FY2024, and was forecast to do the same in Q1 of FY 2025. In contrast, the Chinese market had made up between 20 and 25 percent of the company’s revenue in recent years, until the export restrictions landed.

Read more of this story at Slashdot.