Berlin Bassline Brief #1: Fragnesia, ChatGPT to the bin, RLMs, enclaves
Linux root escalation, XProtect flagging ChatGPT, Recursive Language Models, SSH key protection via Secure Enclave, Mythos finds a curl CVE, and the Secure Enclave explained.
"I’ll walk where my own nature would be leading:
It vexes me to choose another guide."
Security, General
Another week, another Linux root privilege escalation vulnerability. Fragnesia is a variation of the Dirty Frag vulnerability class which has already been making May 2026 so exciting: https://www.bleepingcomputer.com/news/security/new-fragnesia-linux-flaw-lets-attackers-gain-root-privileges/
Security, Apple platforms
XProtect wants you to throw the ChatGPT app out; regrettably, this advice is limited to specific old builds: https://www.heise.de/en/news/Alleged-malware-macOS-sometimes-throws-ChatGPT-into-the-trash-11293967.html
Interesting Paper
Alex L. Zhang, Tim Kraska, Omar Khattab present "Recursive Language Models" – also works with offline-only workflows in my experiments. Accompanying repo will get you going with your own experiments: https://arxiv.org/abs/2512.24601
Interesting Tool
Secretive from Max Goedjen protects your SSH keys with your Mac's Secure Enclave: https://github.com/maxgoedjen/secretive
The weekly "is Mythos real?"
Yes, because security-tuned LLMs are generally good at known vulnerability patterns, and Mythos is one, so if it's the last one you use, the pickings may be slim: https://daniel.haxx.se/blog/2026/05/11/mythos-finds-a-curl-vulnerability/
Apple Platforms Security Concept of the Week
The Secure Enclave: https://support.apple.com/guide/security/the-secure-enclave