KP to Form Own Crisis Management Policy Amid Inflation
April 5, 2026
A major Claude code leak reveals Anthropic’s experimental AI features, including memory systems, autonomous agents and hidden tools.
A major code leak linked to Anthropic has revealed details about upcoming features and experimental tools within its Claude AI ecosystem.
The leaked Claude Code base reportedly spans over 512,000 lines across more than 2,000 files, offering insight into future capabilities under development.
Among the most notable features is “Kairos,” a background system designed to run persistently, enabling proactive task execution and continuous monitoring. The system uses prompts and a file-based memory structure to retain user context across sessions.
Another feature, “AutoDream,” allows the AI to process and refine stored data during idle periods by analysing conversations, removing redundancies and updating information.
The leak also references an “Undercover mode,” which enables AI agents to contribute to open-source repositories anonymously, without revealing their origin.
Additional tools include “UltraPlan” for extended task planning, “Voice Mode” for conversational interaction, and “Bridge Mode” for remote access. A “Coordinator Tool” is also being developed to manage multiple AI agents simultaneously.
The disclosures highlight Anthropic’s focus on autonomous, memory-driven AI systems and advanced enterprise functionality.
