Berlin Bassline Brief #3: Mythos for the many? corecrypto, deceptive LLMs, Poutine, Mach-O
Mythos coming to Claude Code? Formal verification of Apple corecrypto, LLM deception and betrayal, Poutine for CI/CD SAST, and the Mach-O file format explained.
"Gravitation is not responsible for people falling in love."
Security, General
Mythos may be coming to a Claude Code near you; whether this is true exactly as reported or not, the likelihood that a cloud LLM provider opens more vulnerability-chaining capabilities to more users without getting the guardrails perfect from the start makes this a good time for SME/SMB teams to review their planning for a higher volume of incoming issues requiring remediation in 2026: https://www.bleepingcomputer.com/news/artificial-intelligence/anthropics-restricted-claude-mythos-model-may-be-coming-to-claude-code/
Security, Apple platforms
A blueprint for formal verification of Apple corecrypto, by Apple SEAR and Hardware Technologies: https://security.apple.com/blog/formal-verification-corecrypto/ with GitHub repo: https://github.com/apple/corecrypto
Interesting Paper
"A diagnostic hierarchy of epistemic betrayal in large language models" by Danielle Ryder has everything I like in a paper! A multidisciplinary topic where computer science meets philosophy, a framework for categorizing modes of deception, a single author, open access: https://link.springer.com/article/10.1007/s44163-026-01438-2
Interesting Tool
I was originally going to share Boost Security's "Smoked Meat" CI/CD red-teaming tool, but then I realized it was probably more constructive to begin by pointing to their CI/CD SAST Poutine, so you can analyze and remediate your CI/CD workflows before you arrange to attack them: https://github.com/boostsecurityio/poutine
Apple Platforms Security Concept of the Week
The Mach-O file format, by Jonathan Bar Or: https://github.com/yo-yo-yo-jbo/macho_structure