An arXiv abstract credits "the Odin Automatic AI Research Agent" with discovering the proof.

arXiv 2609.11189, "Vector Balancing via Directional Total Variation" by Shengtao Guo, Ethan X. Fang and Junwei Lu, submitted September 10, establishes a 3√(2π) bound for the Komlós signing problem and gives set-system coloring results matching Beck-Fiala predictions. The abstract states plainly that "the proof was discovered by the Odin Automatic AI Research Agent," and says nothing about what Odin is, who operates it, or what it did. 54 upvotes on r/singularity. Unexamined primary document, and the attribution norm question it raises is bigger than the paper.