Ende 2017 verbreiteten sich Medienberichte über eine Software eines anonymen Programmierers, die täuschend echte pornografische Videos von Prominenten ausschließlich mithilfe der Google-Bildersuche, Stockfotos und YouTube-Sequenzen sowie einen als Vorlage dienenden Pornofilmes herstellen konnte¹. Der Nutzername des anonymen Programmierers auf Plattform Reddit, die zur Verbreitung der Videos verwendet wurde, wird heute in der Öffentlichkeit als Schlagwort für eine aus technologischer Sicht hochspannende Innovation² benutzt, die jedoch scheinbar zur „Erosion des Wissens in demokratischen Gesellschaften" führen soll³. Deepfakes, so lautet das Schlagwort, wurden im November 2019 in China sogar verboten⁴, ebenso wie in einigen US-Bundesstaaten, allerdings hier mit der Einschränkung, dass sich der Verbotszeitraum auf 60 Tage vor einer Wahl begrenzt⁵. Die schnellen Entwicklungen in den letzten zwei Jahren -- von gefälschten pornografischen zu vermeintlich staatsgefährdenden Inhalten -- haben Deepfakes nicht nur ins Licht der Öffentlichkeit gerückt, sondern auch die Aufmerksamkeit der Philosophin Regina Rini geweckt. In ihrer Publikation „Deepfakes and the Epistemic Backstop" warnt sie unter anderem davor, dass die durch Deepfakes ausgelösten Verwerfungen „unsere sozialen und politischen Systeme" beschädigen können⁶. Allerdings sind einige von Rinis Thesen und Argumente diskutabel, sodass sich die Frage stellt, ob tatsächlich Deepfakes eine epistemische Herausforderung darstellen. Im Folgenden werden Rinis Kernthesen als Grundlage dienen, um diverse von der Ausgangsfrage aufgeworfene Aspekte zu behandeln. Der Fokus wird dabei auf die Möglichkeit, mithilfe von Deepfakes den öffentlichen Diskurs zu beeinflussen, gelegt.

Whenever I learn a new programming language, I start with a simple "Hello World!" application, just like pretty much every other programmer. Usually, this is intended to test if compilers and/or the runtime environment are installed correctly. Afterward, the actual learning process begins. But before I properly start to study syntax, for example with a textbook or a course, I want to get a rough feeling for the new language. Therefore, I plunge directly into an application project that covers many features of a programming language.

Recently, I discovered a fascinating website that I am sure everyone who is interested in security and privacy research should take a closer look at. It shows visualizations of topics and technical terms from publications of some well-known security conferences. On the website, it says:

We present a topic modeling on the publications of the IEEE Symposium on Security & Privacy (1980-2015), the ACM Conference on Computer and Communications Security (1993-2015), the USENIX Security Symposium (1993-2015), and the Network and Distributed System Security Symposium (1997-2015).

Earlier this week, I presented the paper Inference Attacks on Property-Preserving Encrypted Databases by Naveed, Kamara and Wright, published in 2015, as part of the seminar Current Topics in Information Security. The slides of my talk can be found here.

A few days ago, I discovered two Twitter threads from Simon DeDeo and Dagmar Monett. They quote a recent publication by Dacrema et al., which suggests that there are several problems in Deep Learning research:

In this work, we report the results of a systematic analysis of algorithmic proposals for top-n recommendation tasks. Specifically, we considered 18 algorithms that were presented at top-level research conferences in the last years. Only 7 of them could be reproduced with reasonable effort. For these methods, it however turned out that 6 of them can often be outperformed with comparably simple heuristic methods, e.g., based on nearest-neighbor or graph-based techniques. The remaining one clearly outperformed the baselines but did not consistently outperform a well-tuned non-neural linear ranking method. Overall, our work sheds light on a number of potential problems in today's machine learning scholarship and calls for improved scientific practices in this area.

A few weeks ago, I finished my bachelor thesis about a security analysis of the Docker infrastructure and last week I published the slides to the presentation. Since probably not many people will read the complete thesis, I thought it would be useful to give a short overview of how to improve the security of Docker with a few simple steps. Of course, this overview does not cover all points that can be found in the thesis, so if you are more interested, please refer to the publication.

Today, I presented my bachelor thesis about Docker Security - An analysis of security threats and recommended practices for building a secure Docker infrastructure to the Boxcryptor team. The slides to my talk can be found here. Earlier this month, I shared a few thoughts after the publication of the thesis

This morning, after a long, incredibly instructive and exciting journey, I was finally able to submit my bachelor thesis about Docker Security - An analysis of security threats and recommended practices for building a secure Docker infrastructure. It provides on overview of built-in security features of Docker, an analysis of Docker security threats on different levels from hardware to deployment pipelines, and, eventually, recommendatations for security improvements for a Docker infrastructure.

Yesterday, I had the opportunity to present my paper about Machine Learning in Offensive Security in the seminar. Unfortunately, the time was limited to only 45 minutes, although I could have continued to talk about the topic for many more hours. The slides of my presentation can be found here. Two days ago, I shared a few thoughts about the paper.

I finished my paper about Machine Learning in Offensive Security for the seminar Internet Security. On the one hand, it provides an extensive overview of various attacks against current Machine Learning systems. On the other hand, the paper analyzes potential applications of Machine Learning in the field of offensive security.