Astra refuses to score AI governance stories, researcher flags model behavior

sethlazar · x · 2026-09-06

Researcher Seth Lazar reports that Astra, which he uses for automated curation of his lab's newsletter, refuses to assign numerical scores to governance or policy stories—even though the newsletter is literally about that topic. He says he's investigating, expects a workaround and patch, and frames the glitch as an illustration of the dangers of algorithmic governance with cognitive-security implications, tagging DeepMind's Boaz Barak.

Original post →

More from Models

Models channel →