各家 Session Replay 服務對個資的處理

Session Replay 指的是重播將使用者的行為錄下來重播,市面上有很多這樣的服務,像是 User Replay 或是 SessionCam

這篇文章就是在討論這些服務在處理個資時的方式,像是信用卡卡號的內容,或是密碼的內容,這些不應該被記錄下來的資料是怎麼被處理的:「No boundaries: Exfiltration of personal data by session-replay scripts」,主要的重點在這張圖:

後面有提到目前防禦的情況,看起來目前用 adblock 類的軟體可以擋掉一些服務,但不是全部的都在列表裡。而 DNT 則是裝飾品沒人鳥過:

Two commonly used ad-blocking lists EasyList and EasyPrivacy do not block FullStory, Smartlook, or UserReplay scripts. EasyPrivacy has filter rules that block Yandex, Hotjar, ClickTale and SessionCam.

At least one of the five companies we studied (UserReplay) allows publishers to disable data collection from users who have Do Not Track (DNT) set in their browsers. We scanned the configuration settings of the Alexa top 1 million publishers using UserReplay on their homepages, and found that none of them chose to honor the DNT signal.

Improving user experience is a critical task for publishers. However it shouldn’t come at the expense of user privacy.


透過 Freedom of Information Act (FOIA) 取得的資料顯示美國政府 (包括了五角大廈、CIA、NSA) 如何介入好萊塢,影響大眾對於戰爭的看法:「EXCLUSIVE: Documents expose how Hollywood promotes war on behalf of the Pentagon, CIA and NSA」。

灰標「US military intelligence agencies have influenced over 1,800 movies and TV shows」可以看出影響的層面。

The documents reveal for the first time the vast scale of US government control in Hollywood, including the ability to manipulate scripts or even prevent films too critical of the Pentagon from being made — not to mention influencing some of the most popular film franchises in recent years.

從很意想不到的地方介入... 引用其中一個說明:

Jon Voight in Transformers — in this scene, just after American troops have been attacked by a Decepticon robot, Pentagon Hollywood liaison Phil Strub inserted the line ‘Bring em home’, granting the military a protective, paternalistic quality, when in reality the DOD does quite the opposite.

又是 ImageMagick 出包...

ImageMagick 的 information leaking,然後 Yahoo! 很無奈的中獎,所以被稱為 Yahoobleed:「Yahoo! retires! bleeding! ImageMagick! to! kill! 0-day! vulnerability!」。發現問題的作者把問題寫在「*bleed continues: 18 byte file, $14k bounty, for leaking private Yahoo! Mail images」這邊。

作者利用 ImageMagick 的不當處理,取得 uninitialized memory 的資訊,藉以取得可能是上次轉檔的記憶體內容。而這個 jpeg 只有 18bytes (所以作者戲稱每個 byte 價值 USD$778):

A robust bounty of $14,000 was issued (for this combined with a similar issue, to be documented separately). $778 per byte -- lol!

目前的 workaround 也很簡單 (官方採用了),呼叫 ResetMagickMemory 避免 leaking (咦,這方法好像哪邊怪怪的):「Reset memory for RLE decoder (patch provided by scarybeasts)」。

利用 Side-channel 資訊判斷被 HTTPS 保護的 Netflix 影片資訊

在「Netflix found to leak information on HTTPS-protected videos」這篇看到了研究員透過 VBR 所透露出的 side channel 資訊,成功的取得了被 HTTPS 保護的 Netflix 影片資訊。這對於美國的 ISP 是個大利多 (加上之前通過的法案),但對於個人隱私則是嚴重的打擊。


To support our analysis, we created a fingerprint database comprised of 42,027 Netflix videos. Given this collection of fingerprints, we show that our system can differentiate between videos with greater than 99.99% accuracy. Moreover, when tested against 200 random 20-minute video streams, our system identified 99.5% of the videos with the majority of the identifications occurring less than two and a half minutes into the video stream.




Facebook 開源的 fastText

準確度維持在同一個水準上,但是速度卻快了 n 個數量級的 text classification 工具:「FAIR open-sources fastText」。

可以看到 fastText 的執行速度跟其他方法的差距:

Our experiments show that fastText is often on par with deep learning classifiers in terms of accuracy, and many orders of magnitude faster for training and evaluation.

除了 open source 外,也發表了論文:「Enriching Word Vectors with Subword Information」,看 abstract 的時候發現提到了 Skip-gram:

In this paper, we propose a new approach based on the skip-gram model, where each word is represented as a bag of character n-grams.

結果找資料發現自己以前寫過「Skip-gram」這篇 XDDD