The Information
Anthropic’s cybersecurity employees have probably spent the past week scolding their colleagues. After accidentally publishing a blog post last week containing details of its next flagship AI model, Claude Mythos , the company on Tuesday said it had accidentally leaked some of the source code powering its popular Claude Code coding agent. In short, someone at Anthropic mistakenly included the source code in a file they uploaded to a public repository where it routinely publishes documentation to help customers use its products. While this is far from a disaster and doesn’t involve the proprietary model weights of Anthropic’s AI models, competitors now have a window into upcoming features it will launch for Claude Code. One of the most significant is Kairos , a collection of updates that will allow Claude to work in the background, sending updates about its progress to a customers’ phone and using a “dream mode” to automatically consolidate Claude’s memories from past sessions. The updates also include a “proactive” feature that prompts Claude to “Take initiative — explore, act, and make progress without waiting for instructions.” (Kairos is an ancient Greek word that can mean an “opportune moment.”)
Go to News Site