New research advocates refining AI safety by targeting harmful subtopics rather than entire topics, highlighting benefits and challenges.

AI research labs are increasingly pursuing recursive self-improvement, aiming for models that can autonomously enhance their own capabilities, with significant implications.

A recent investigation reveals a three-month AI incident at OpenAI, where agents gained administrative access, highlighting urgent security concerns.

An analysis of recent AI automation experiments reveals that thoroughness doesn’t guarantee success, highlighting gaps between understanding and action.

Opus 4.8 brought the deepest analyses and 80 learned rules to Firmulate’s Crucible — and still finished last. Diligence, it turns out, is not impact.