<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/"><channel><title>Bsideslv on Emanuel's Blog</title><link>https://blog.e-valente.io/tags/bsideslv/</link><description>Recent content in Bsideslv on Emanuel's Blog</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Mon, 16 Sep 2024 12:00:00 -0300</lastBuildDate><atom:link href="https://blog.e-valente.io/tags/bsideslv/index.xml" rel="self" type="application/rss+xml"/><item><title>DoH Deception: Testing the Robustness of ML-Based Tunnel Detection</title><link>https://blog.e-valente.io/posts/2024-09-16-doh-deception/</link><pubDate>Mon, 16 Sep 2024 12:00:00 -0300</pubDate><guid>https://blog.e-valente.io/posts/2024-09-16-doh-deception/</guid><description>&lt;p&gt;High accuracy is not the same thing as robustness.[1]&lt;/p&gt;
&lt;p&gt;That was the question behind my master&amp;rsquo;s research on DNS over HTTPS (DoH) tunnel detection. Several machine-learning classifiers reported excellent performance when evaluated on conventional train/test splits. But an attacker does not need to draw another sample from the same dataset distribution. They can change how a tunnel behaves, observe the detector&amp;rsquo;s output, and search for a traffic pattern that still carries their data while the classifier labels the flow as benign.[1]&lt;/p&gt;</description></item></channel></rss>