
also remains robust under obfuscation (F1 above 0.90 for 7/8 Obfuscapk operators where baselines drop sharply) and temporal concept drift (F1 of 0.99 on 2022 and 0.95 on 2023 apps).

Software Engineering and Security Group at Umeå University, Sweden.

also remains robust under obfuscation (F1 above 0.90 for 7/8 Obfuscapk operators where baselines drop sharply) and temporal concept drift (F1 of 0.99 on 2022 and 0.95 on 2023 apps).

... on the selected tokens. Evaluated on a cleaned public benchmark (PublicAMD) and a new AndroZoo-derived dataset (AndroAMD), Loom reaches F1 =0.97 on both, outperforming all string-, image-, graph-, and transformer-based baselines.

In this work, we propose Loom, a transformer-based detector that operates directly on filtered string-based tokens extracted from the AndroidManifest.xml, Smali API calls, and Smali opcode strings. We compare seven classical filter methods and fine-tune BERTBASE ...

Our paper "Loom: A Balanced String-Based Transformer for Android Malware Detection" will be presented at the 28th International Conference on Information and Communications Security (ICICS 2026) in October!


I am very happy to receive Nordea's Scientific Prize 2026! It is the result of many research and teaching collaborations, and especially of the hard-working group members I have the pleasure of working with. Many thanks to all of you and have a great summer ;-) !

[1] Ken Thompson, "Reflections on trusting trust." Communications of the ACM 27.8 (1984): 761-763.

Thanks to Mattias Wadenstein for the invitation and the organisation of this event! For more details about these two attacks, you can visit the Github pages: github.com/software-engineer… github.com/software-engineer…

We are going back to 1984: "No amount of source-level verification or scrutiny will protect you from using untrusted code" [1] s/source/bytecode/.

Then, I explained how bugs in the Java or Android VMs could be used to hide the real behavior of the code from human analysts or tools. This could be used to bypass some vetting mechanisms of app stores such as Google Play.

Since these code changes are very small and look benign, it is very challenging for an analyst to understand what the security consequences are.

First, an attacker could inject weaknesses or vulnerabilities in the code via very tiny and stealthy additions of a few keywords. In the context of Java deserialization vulnerabilities, this allows an attacker to enable complete gadget chains in popular Java libraries.

On Tuesday, I gave a keynote at BSides Ume on "Software Supply Chain Attacks" where I presented two examples of what could go wrong in the software supply chain.

My article on hidden messages within GPS signals was featured on @404mediaco. 404media.co/the-u-s-military…

Thank you to Chatarina Larson for organizing this event!

Another interesting aspect that was mentioned is the homogeneity of software used at universities such as software handling emails. If a vulnerability (technical or legal) is found and exploited it could expose data for all institutions relying on this software.