Contexte

These sources document a turbulent period for Anthropic in early 2026, highlighted by the accidental exposure of the Claude Code source code through a misconfigured npm file. While the leak provided a detailed architectural blueprint for AI agents, it also allowed researchers to identify a critical vulnerability in the tool’s permission system. Simultaneously, documentation from Amazon Bedrock and official release notes reveal a rapid expansion of the Claude ecosystem, including the debut of the high-intelligence Claude Mythos model and the general availability of Claude Cowork. The updates further introduce advanced autonomous capabilities, such as computer use and persistent agent threads, designed to integrate AI deeper into professional workflows. Technical repositories also show how these tools interact with standardized protocols like the Model Context Protocol (MCP) to manage external data and robotic interfaces. Ultimately, the collection portrays an industry leader balancing aggressive feature deployment with significant security and oversight challenges.

Sources

Voir les 4 sources restantes