Topic
10 written stories · 17 tracked
Claude Code will enable auto mode by default for its Pro, Max, and Team plans, effective August 14. The change is intended to help detect harmful actions automatically. The update applies to all users of the three paid tiers.
Alibaba has prohibited its employees from using Claude Code, directing them to remove all Claude models from their work computers. The ban was implemented over alleged security concerns, including potential backdoor risks and hidden tracking of Chinese users. The company’s action reflects heightened caution regarding the AI tool’s safety.
Z.ai has launched ZCode, an “Agentic Development Environment” built to work with its new GLM-5.2 model. The platform is positioned as a competitor to existing AI-assisted coding tools such as Cursor, Claude Code, and GitHub Copilot, offering developers a dedicated environment for generating and managing code with the latest large language model.
Recent analyses highlight that conventional AI benchmarks often fail to reflect real-world performance, overlooking practical constraints such as latency and resource usage. In response, a newly introduced optimization framework has demonstrated a 2.5-fold efficiency gain over leading models like Claude Code and Codex while operating within the same compute budget. This development underscores growing concerns about the gap between benchmark results and actual deployment effectiveness.
Xiaomi’s MiMo AI team has launched MiMo Code V0.1.0, a terminal-native AI coding assistant that is now open-source. The tool is reported to outperform Claude Code on agentic coding and software-engineering benchmarks, including ultra-long tasks that exceed 200 steps.
Anthropic announced that it is increasing the usage limits for its Claude Code model, doubling the rate limits for paid plans and eliminating the peak-hour reduction that previously applied to Pro and Max subscriptions. The company attributes the expanded limits to a new compute agreement with SpaceX.
A misconfigured npm package exposed Claude Code’s full TypeScript source, revealing internal codenames, a “Self-Healing Memory” architecture, and new Kairos updates that let Claude run in the background and use a “dream mode” to consolidate memories. Anthropic said the leak was a packaging error caused by human mistake, not a security breach.
Anthropic has released a safer “auto mode” for its Claude Code and Cowork tools, enabling the AI to make permission-level decisions while blocking destructive actions such as mass file deletion. The feature allows the models to control a user’s computer, providing a research preview on macOS for Pro and Max subscribers.
Anthropic has introduced a code-review feature for its Claude Code model that uses multiple agents working together to examine pull requests for bugs. The tool is being offered as a research preview, aimed at handling the growing volume of AI-generated code.
Anthropic has expanded its enterprise offerings by launching Claude Cowork, a suite of agent tools designed for sectors including finance, HR, and design. These tools integrate with various enterprise applications to automate multi-step tasks. Additionally, the company introduced Remote Control for Claude Code, a feature that allows users to manage terminal sessions through mobile devices and web interfaces.