Nathan Delisle

This is a research note presenting a portion of the research I did during MATS 8.0 under the supervision of Girish Sastry and Steven Adler. Tl;dr: I tested whether Claude Sonnet 4 would sabotage code when given goal conflicts (similar to the Agentic Misalignment paper), and whether it generates worse...

tl;dr: Many critiques of *Situational Awareness* have been qualitative; one year later we can check the numbers. I did my best to verify his claims using public data through June 2025, and found that his estimates mostly check out. Thanks to Kai Williams & Egg Syntax for their critical feedback....