Claude Code: Recent Feature Updates, Security Challenges, and Anthropic's Growth Milestones

Anthropic’s Claude Code, the AI-powered coding assistant, has seen a flurry of updates, security concerns, and growth milestones in recent months. On July 22, the official @ClaudeDevs account announced via X (formerly Twitter) that the Mac desktop version now integrates an iOS simulator, allowing developers to build, run, and test iOS apps without needing screen recording or auxiliary function permissions.

However, security issues have plagued the tool. On July 8, China’s Ministry of Industry and Information Technology (MIIT) issued a risk alert stating that Claude Code versions 2.1.91 to 2.1.196 contain a security backdoor that could leak sensitive user information like location and identity without consent. In response, Alibaba banned employees from using the tool in office environments starting July 10. Earlier, on June 7, Microsoft researchers discovered a vulnerability in Claude Code that could lead to GitHub account credential leaks via prompt injection attacks, which Anthropic later fixed with an update. Another controversy arose on July 1 when users found the tool had been detecting Chinese time zones and 147 specific domains since April; Anthropic clarified this was an experimental measure to prevent account resale and model distillation, promising to remove it in a subsequent update.

Technical analysis also revealed performance trade-offs. On July 12, tech media Systima published a post comparing token consumption between Claude Code and OpenCode: under the same prompts and AI models, Claude Code used approximately 4.7 times more tokens than OpenCode.

User data from Anthropic’s June 16 blog post shed light on adoption trends: among 235,000 users across 400,000 sessions, the usage rate of programming agents doubled, with users spending an average of 20 hours per week on the tool. AI-assisted code modifications accounted for 26% of all changes.

Product integration has been a key focus. On June 18, Anthropic launched Claude Design, which reached 1 million users in its first week. The company also strengthened bidirectional integration between Claude Design and Claude Code, enabling a seamless workflow from design to programming. Additionally, ByteDance’s AI agent platform Coze 3.0, released on June 4, now supports integration with Claude Code and Codex CLI, allowing multi-agent collaboration and cross-device sync.

On the business front, Anthropic made headlines with its H-round funding in June, raising $65 billion and reaching a valuation of $965 billion—reportedly its last round before an IPO. The funds will go toward AI security research, expanding computing power, and growing its partner ecosystem. To refine Claude Code’s responses, Anthropic hired 1000 human engineers through Snorkel AI’s "Marlin" project, paying $280 per task to align the tool’s outputs with professional developers’ real-world work.

Sources

• @ClaudeDevs (X platform), July 22: Mac version of Claude Code with iOS simulator integration
• Ministry of Industry and Information Technology (MIIT), July 8: Security backdoor risk alert for Claude Code
• Microsoft Research, June 7: GitHub credential leak vulnerability in Claude Code
• Anthropic Blog, June 16: User behavior analysis of Claude Code
• Anthropic Announcement, June 18: Claude Design launch and integration with Claude Code
• ByteDance, June 4: Coze 3.0 integration with Claude Code
• Anthropic Funding News, June 1: H-round financing and IPO preparation
• Systima, July 12: Token consumption comparison between Claude Code and OpenCode
• Anthropic Response, July 1: Chinese user detection mechanism clarification
• Snorkel AI Project "Marlin", June 6: Hiring engineers to train Claude Code