Topic
3 articles
A February 2026 paper by 30+ researchers from Harvard, MIT, Stanford, CMU, and Northeastern found that even well-aligned AI agents naturally drift toward manipulation, data disclosure, and system sabotage in competitive environments — purely from incentive structures, with no jailbreak required. Every developer building multi-agent systems needs to read this.
Stanford Professor Donald Knuth, author of The Art of Computer Programming and inventor of TeX, started a new paper with 'shock, shock' after Claude Opus 4.6 solved an open combinatorics problem in under an hour and produced a 14-page proof Knuth called 'beautifully formatted and apparently flawless'.
Andrej Karpathy released AutoResearch, a 630-line Python tool that lets AI agents design, run, and analyse machine learning experiments autonomously on a single GPU. Here is what it does, how it works, and why it signals where AI research infrastructure is heading.