Anthropic on Thursday urged leading AI labs to consider slowing frontier AI
development, saying rapid model advances could soon enable autonomous
self‑improvement without human intervention and create major societal risk. In a
blog post releasing internal data, Anthropic documented swift performance gains
in its top models and said progress appears to be moving toward “recursive
self‑improvement.” The company proposed a global agreement to slow frontier AI
work and a mechanism to verify competitors’ compliance, arguing a pause would
allow social institutions and related research to catch up.
