Anthropic has unveiled three new features for Claude Code, including an innovative auto mode that leverages an AI safety classifier to evaluate tool calls in real-time. This feature automatically approves routine operations while blocking potentially destructive ones, providing an additional layer of security.

The introduction of auto mode aims to strike a balance between manual approval and the skip permissions flag, offering users a middle ground. However, the classifier’s decision-making process remains unclear, as Anthropic has not publicly disclosed the specific rules governing what is allowed or blocked.

In related news, Anthropic has launched Channels, enabling users to control Claude Code through popular platforms like Discord and Telegram. Furthermore, GitHub has announced plans to train on Copilot user data by default starting April 24, marking a significant development in AI-driven collaboration.

Meanwhile, Cursor has launched Composer 2, but the company has faced controversy over its use of Moonshot AI’s Kimi K2.5 model without proper disclosure. This incident underscores the importance of transparency in AI development and highlights the need for clear guidelines on AI safety and trust.

Photo by Natalia S on Pexels
Photos provided by Pexels