Anthropic reveals Model 2, raises AI risk level

Anthropic has disclosed the existence of two unreleased successors to Claude Mythos 5, called Model 1 and Model 2, in its latest 186-page AI alignment report. Model 2, the more capable of the two, is heavily used internally and offers noticeable improvements over Mythos 5 for engineering tasks like coding and generating training data. The report also raises the risk level for Threat Model 2 situations, covering AI interference with organizational systems, from very low to low, citing recent cybersecurity incidents. Anthropic also flagged growing uncertainty around recursive self-improvement risks, noting its internal benchmarks are struggling to keep pace with rapid LLM advances.